From: Ali Saidi Date: Tue, 30 May 2006 22:57:42 +0000 (-0400) Subject: Add a very poor implementation of dealing with retries on timing requests. It is... X-Git-Tag: m5_2.0_beta1~36^2~116^2^2 X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=09d8a1e1252eba582fd8450ee31e784b54910f7d;p=gem5.git Add a very poor implementation of dealing with retries on timing requests. It is especially slow with tracing on since it ends up being O(N^2). But it's probably going to have to change for the real bus anyway, so it should be rewritten then Change recvRetry() to not accept a packet. Sendtiming should be called again (and can respond with false or true) Removed Port Blocked/Unblocked and replaced with sendRetry(). Remove possibility of packet mangling if packet is going to be refused anyway in bridge src/cpu/simple/atomic.cc: src/cpu/simple/atomic.hh: src/cpu/simple/timing.cc: src/cpu/simple/timing.hh: Change recvRetry() to not accept a packet. Sendtiming should be called again (and can respond with false or true) src/dev/io_device.cc: src/dev/io_device.hh: Make DMA Timing requests/responses work. Change recvRetry() to not accept a packet. Sendtiming should be called again (and can respond with false or true) src/mem/bridge.cc: src/mem/bridge.hh: Change recvRetry() to not accept a packet. Sendtiming should be called again (and can respond with false or true) Removed Port Blocked/Unblocked and replaced with sendRetry(). Remove posibility of packet mangling if packet is going to be refused anyway. src/mem/bus.cc: src/mem/bus.hh: Add a very poor implementation of dealing with retries on timing requests. It is especially slow with tracing on since it ends up being O(N^2). But it's probably going to have to change for the real bus anyway, so it should be rewritten then src/mem/port.hh: Change recvRetry() to not accept a packet. Sendtiming should be called again (and can respond with false or true) Removed Blocked/Unblocked port status, their functionality is really duplicated in the recvRetry() method --HG-- extra : convert_revision : fab613404be54bfa7a4c67572bae7b559169e573 --- diff --git a/src/cpu/simple/atomic.cc b/src/cpu/simple/atomic.cc index 3cad6e43f..a0d26a8ab 100644 --- a/src/cpu/simple/atomic.cc +++ b/src/cpu/simple/atomic.cc @@ -106,11 +106,10 @@ AtomicSimpleCPU::CpuPort::recvStatusChange(Status status) panic("AtomicSimpleCPU doesn't expect recvStatusChange callback!"); } -Packet * +void AtomicSimpleCPU::CpuPort::recvRetry() { panic("AtomicSimpleCPU doesn't expect recvRetry callback!"); - return NULL; } diff --git a/src/cpu/simple/atomic.hh b/src/cpu/simple/atomic.hh index ab3a3e8ef..65269bd6d 100644 --- a/src/cpu/simple/atomic.hh +++ b/src/cpu/simple/atomic.hh @@ -98,7 +98,7 @@ class AtomicSimpleCPU : public BaseSimpleCPU virtual void recvStatusChange(Status status); - virtual Packet *recvRetry(); + virtual void recvRetry(); virtual void getDeviceAddressRanges(AddrRangeList &resp, AddrRangeList &snoop) diff --git a/src/cpu/simple/timing.cc b/src/cpu/simple/timing.cc index 7cdcdafa1..5f094d033 100644 --- a/src/cpu/simple/timing.cc +++ b/src/cpu/simple/timing.cc @@ -419,17 +419,18 @@ TimingSimpleCPU::IcachePort::recvTiming(Packet *pkt) return true; } -Packet * +void TimingSimpleCPU::IcachePort::recvRetry() { // we shouldn't get a retry unless we have a packet that we're // waiting to transmit assert(cpu->ifetch_pkt != NULL); assert(cpu->_status == IcacheRetry); - cpu->_status = IcacheWaitResponse; Packet *tmp = cpu->ifetch_pkt; - cpu->ifetch_pkt = NULL; - return tmp; + if (sendTiming(tmp)) { + cpu->_status = IcacheWaitResponse; + cpu->ifetch_pkt = NULL; + } } void @@ -459,17 +460,18 @@ TimingSimpleCPU::DcachePort::recvTiming(Packet *pkt) return true; } -Packet * +void TimingSimpleCPU::DcachePort::recvRetry() { // we shouldn't get a retry unless we have a packet that we're // waiting to transmit assert(cpu->dcache_pkt != NULL); assert(cpu->_status == DcacheRetry); - cpu->_status = DcacheWaitResponse; Packet *tmp = cpu->dcache_pkt; - cpu->dcache_pkt = NULL; - return tmp; + if (sendTiming(tmp)) { + cpu->_status = DcacheWaitResponse; + cpu->dcache_pkt = NULL; + } } diff --git a/src/cpu/simple/timing.hh b/src/cpu/simple/timing.hh index b46631d5a..cb37824bc 100644 --- a/src/cpu/simple/timing.hh +++ b/src/cpu/simple/timing.hh @@ -100,7 +100,7 @@ class TimingSimpleCPU : public BaseSimpleCPU virtual bool recvTiming(Packet *pkt); - virtual Packet *recvRetry(); + virtual void recvRetry(); }; class DcachePort : public CpuPort @@ -115,7 +115,7 @@ class TimingSimpleCPU : public BaseSimpleCPU virtual bool recvTiming(Packet *pkt); - virtual Packet *recvRetry(); + virtual void recvRetry(); }; IcachePort icachePort; diff --git a/src/dev/io_device.cc b/src/dev/io_device.cc index 87fdbb765..563fdb759 100644 --- a/src/dev/io_device.cc +++ b/src/dev/io_device.cc @@ -26,6 +26,7 @@ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ +#include "base/trace.hh" #include "dev/io_device.hh" #include "sim/builder.hh" @@ -55,12 +56,13 @@ PioPort::getDeviceAddressRanges(AddrRangeList &resp, AddrRangeList &snoop) } -Packet * +void PioPort::recvRetry() { Packet* pkt = transmitList.front(); - transmitList.pop_front(); - return pkt; + if (Port::sendTiming(pkt)) { + transmitList.pop_front(); + } } @@ -74,6 +76,7 @@ PioPort::SendEvent::process() } + bool PioPort::recvTiming(Packet *pkt) { @@ -116,16 +119,20 @@ DmaPort::recvTiming(Packet *pkt) { if (pkt->senderState) { DmaReqState *state; + DPRINTF(DMA, "Received response Packet %#x with senderState: %#x\n", + pkt, pkt->senderState); state = dynamic_cast(pkt->senderState); - state->completionEvent->schedule(pkt->time - pkt->req->getTime()); + assert(state); + state->completionEvent->process(); delete pkt->req; delete pkt; } else { + DPRINTF(DMA, "Received response Packet %#x with no senderState\n", pkt); delete pkt->req; delete pkt; } - return Packet::Success; + return true; } DmaDevice::DmaDevice(Params *p) @@ -133,20 +140,19 @@ DmaDevice::DmaDevice(Params *p) { } void -DmaPort::SendEvent::process() -{ - if (port->Port::sendTiming(packet)) - return; - - port->transmitList.push_back(packet); -} - -Packet * DmaPort::recvRetry() { Packet* pkt = transmitList.front(); - transmitList.pop_front(); - return pkt; + DPRINTF(DMA, "Retry on Packet %#x with senderState: %#x\n", + pkt, pkt->senderState); + if (sendTiming(pkt)) { + DPRINTF(DMA, "-- Done\n"); + transmitList.pop_front(); + pendingCount--; + assert(pendingCount >= 0); + } else { + DPRINTF(DMA, "-- Failed, queued\n"); + } } @@ -192,13 +198,22 @@ DmaPort::sendDma(Packet *pkt) // switching actually work /* MemState state = device->platform->system->memState; - if (state == Timing) { - if (!sendTiming(pkt)) - transmitList.push_back(&packet); - } else if (state == Atomic) {*/ + if (state == Timing) { */ + DPRINTF(DMA, "Attempting to send Packet %#x with senderState: %#x\n", + pkt, pkt->senderState); + if (!sendTiming(pkt)) { + transmitList.push_back(pkt); + DPRINTF(DMA, "-- Failed: queued\n"); + } else { + DPRINTF(DMA, "-- Done\n"); + pendingCount--; + assert(pendingCount >= 0); + } + /* } else if (state == Atomic) { sendAtomic(pkt); if (pkt->senderState) { DmaReqState *state = dynamic_cast(pkt->senderState); + assert(state); state->completionEvent->schedule(curTick + (pkt->time - pkt->req->getTime()) +1); } pendingCount--; @@ -206,7 +221,7 @@ DmaPort::sendDma(Packet *pkt) delete pkt->req; delete pkt; -/* } else if (state == Functional) { + } else if (state == Functional) { sendFunctional(pkt); // Is this correct??? completionEvent->schedule(pkt->req->responseTime - pkt->req->requestTime); diff --git a/src/dev/io_device.hh b/src/dev/io_device.hh index 74730ad92..5836da3c9 100644 --- a/src/dev/io_device.hh +++ b/src/dev/io_device.hh @@ -105,8 +105,9 @@ class PioPort : public Port void sendTiming(Packet *pkt, Tick time) { new PioPort::SendEvent(this, pkt, time); } - /** This function pops the last element off the transmit list and sends it.*/ - virtual Packet *recvRetry(); + /** This function is notification that the device should attempt to send a + * packet again. */ + virtual void recvRetry(); public: PioPort(PioDevice *dev, Platform *p); @@ -146,28 +147,11 @@ class DmaPort : public Port virtual void recvStatusChange(Status status) { ; } - virtual Packet *recvRetry() ; + virtual void recvRetry() ; virtual void getDeviceAddressRanges(AddrRangeList &resp, AddrRangeList &snoop) { resp.clear(); snoop.clear(); } - class SendEvent : public Event - { - DmaPort *port; - Packet *packet; - - SendEvent(PioPort *p, Packet *pkt, Tick t) - : Event(&mainEventQueue), packet(pkt) - { schedule(curTick + t); } - - virtual void process(); - - virtual const char *description() - { return "Future scheduled sendTiming event"; } - - friend class DmaPort; - }; - void sendDma(Packet *pkt); public: @@ -178,8 +162,6 @@ class DmaPort : public Port bool dmaPending() { return pendingCount > 0; } - friend class DmaPort::SendEvent; - }; /** diff --git a/src/mem/bridge.cc b/src/mem/bridge.cc index 736e8dc81..5b8b9164b 100644 --- a/src/mem/bridge.cc +++ b/src/mem/bridge.cc @@ -90,7 +90,17 @@ Bridge::BridgePort::recvTiming(Packet *pkt) DPRINTF(BusBridge, "recvTiming: src %d dest %d addr 0x%x\n", pkt->getSrc(), pkt->getDest(), pkt->getAddr()); - if (pkt->isResponse()) { + return otherPort->queueForSendTiming(pkt); +} + + +bool +Bridge::BridgePort::queueForSendTiming(Packet *pkt) +{ + if (queueFull()) + return false; + + if (pkt->isResponse()) { // This is a response for a request we forwarded earlier. The // corresponding PacketBuffer should be stored in the packet's // senderState field. @@ -99,22 +109,16 @@ Bridge::BridgePort::recvTiming(Packet *pkt) // set up new packet dest & senderState based on values saved // from original request buf->fixResponse(pkt); + DPRINTF(BusBridge, "restoring sender state: %#X, from packet buffer: %#X\n", + pkt->senderState, buf); DPRINTF(BusBridge, " is response, new dest %d\n", pkt->getDest()); delete buf; } - return otherPort->queueForSendTiming(pkt); -} - - -bool -Bridge::BridgePort::queueForSendTiming(Packet *pkt) -{ - if (queueFull()) - return false; - Tick readyTime = curTick + delay; PacketBuffer *buf = new PacketBuffer(pkt, readyTime); + DPRINTF(BusBridge, "old sender state: %#X, new sender state: %#X\n", + buf->origSenderState, buf); // If we're about to put this packet at the head of the queue, we // need to schedule an event to do the transmit. Otherwise there @@ -126,43 +130,16 @@ Bridge::BridgePort::queueForSendTiming(Packet *pkt) sendQueue.push_back(buf); - // Did we just become blocked? If yes, let other side know. - if (queueFull()) - otherPort->sendStatusChange(Port::Blocked); - return true; } - -void -Bridge::BridgePort::finishSend(PacketBuffer *buf) -{ - if (buf->expectResponse) { - // Must wait for response. We just need to count outstanding - // responses (in case we want to cap them); PacketBuffer - // pointer will be recovered on response. - ++outstandingResponses; - DPRINTF(BusBridge, " successful: awaiting response (%d)\n", - outstandingResponses); - } else { - // no response expected... deallocate packet buffer now. - DPRINTF(BusBridge, " successful: no response expected\n"); - delete buf; - } - - // If there are more packets to send, schedule event to try again. - if (!sendQueue.empty()) { - buf = sendQueue.front(); - sendEvent.schedule(std::max(buf->ready, curTick + 1)); - } -} - - void Bridge::BridgePort::trySend() { assert(!sendQueue.empty()); + bool was_full = queueFull(); + PacketBuffer *buf = sendQueue.front(); assert(buf->ready <= curTick); @@ -176,20 +153,41 @@ Bridge::BridgePort::trySend() // send successful sendQueue.pop_front(); buf->pkt = NULL; // we no longer own packet, so it's not safe to look at it - finishSend(buf); + + if (buf->expectResponse) { + // Must wait for response. We just need to count outstanding + // responses (in case we want to cap them); PacketBuffer + // pointer will be recovered on response. + ++outstandingResponses; + DPRINTF(BusBridge, " successful: awaiting response (%d)\n", + outstandingResponses); + } else { + // no response expected... deallocate packet buffer now. + DPRINTF(BusBridge, " successful: no response expected\n"); + delete buf; + } + + // If there are more packets to send, schedule event to try again. + if (!sendQueue.empty()) { + buf = sendQueue.front(); + sendEvent.schedule(std::max(buf->ready, curTick + 1)); + } + // Let things start sending again + if (was_full) { + DPRINTF(BusBridge, "Queue was full, sending retry\n"); + otherPort->sendRetry(); + } + } else { DPRINTF(BusBridge, " unsuccessful\n"); } } -Packet * +void Bridge::BridgePort::recvRetry() { - PacketBuffer *buf = sendQueue.front(); - Packet *pkt = buf->pkt; - finishSend(buf); - return pkt; + trySend(); } /** Function called by the port when the bus is receiving a Atomic @@ -223,9 +221,6 @@ Bridge::BridgePort::recvFunctional(Packet *pkt) void Bridge::BridgePort::recvStatusChange(Port::Status status) { - if (status == Port::Blocked || status == Port::Unblocked) - return; - otherPort->sendStatusChange(status); } diff --git a/src/mem/bridge.hh b/src/mem/bridge.hh index 8a5cbf92a..8c7af562c 100644 --- a/src/mem/bridge.hh +++ b/src/mem/bridge.hh @@ -38,7 +38,6 @@ #include #include - #include "mem/mem_object.hh" #include "mem/packet.hh" #include "mem/port.hh" @@ -77,7 +76,8 @@ class Bridge : public MemObject origSenderState(_pkt->senderState), origSrc(_pkt->getSrc()), expectResponse(_pkt->needsResponse()) { - pkt->senderState = this; + if (!pkt->isResponse()) + pkt->senderState = this; } void fixResponse(Packet *pkt) @@ -146,7 +146,7 @@ class Bridge : public MemObject /** When receiving a retry request from the peer port, pass it to the bridge. */ - virtual Packet* recvRetry(); + virtual void recvRetry(); /** When receiving a Atomic requestfrom the peer port, pass it to the bridge. */ diff --git a/src/mem/bus.cc b/src/mem/bus.cc index cfc99a64f..aec3ef2f6 100644 --- a/src/mem/bus.cc +++ b/src/mem/bus.cc @@ -72,9 +72,35 @@ Bus::recvTiming(Packet *pkt) assert(dest != pkt->getSrc()); // catch infinite loops port = interfaces[dest]; } - return port->sendTiming(pkt); + if (port->sendTiming(pkt)) { + // packet was successfully sent, just return true. + return true; + } + + // packet not successfully sent + retryList.push_back(interfaces[pkt->getSrc()]); + return false; +} + +void +Bus::recvRetry(int id) +{ + // Go through all the elements on the list calling sendRetry on each + // This is not very efficient at all but it works. Ultimately we should end + // up with something that is more intelligent. + int initialSize = retryList.size(); + int i; + Port *p; + + for (i = 0; i < initialSize; i++) { + assert(retryList.size() > 0); + p = retryList.front(); + retryList.pop_front(); + p->sendRetry(); + } } + Port * Bus::findPort(Addr addr, int id) { diff --git a/src/mem/bus.hh b/src/mem/bus.hh index 5eeb07904..774f9f3a4 100644 --- a/src/mem/bus.hh +++ b/src/mem/bus.hh @@ -67,6 +67,10 @@ class Bus : public MemObject transaction.*/ void recvFunctional(Packet *pkt); + /** Timing function called by port when it is once again able to process + * requests. */ + void recvRetry(int id); + /** Function called by the port when the bus is recieving a status change.*/ void recvStatusChange(Port::Status status, int id); @@ -126,6 +130,11 @@ class Bus : public MemObject virtual void recvStatusChange(Status status) { bus->recvStatusChange(status, id); } + /** When reciving a retry from the peer port (at id), + pass it to the bus. */ + virtual void recvRetry() + { bus->recvRetry(id); } + // This should return all the 'owned' addresses that are // downstream from this bus, yes? That is, the union of all // the 'owned' address ranges of all the other interfaces on @@ -143,6 +152,10 @@ class Bus : public MemObject connected to this bus.*/ std::vector interfaces; + /** An array of pointers to ports that retry should be called on because the + * original send failed for whatever reason.*/ + std::list retryList; + public: /** A function used to return the port associated with this bus object. */ diff --git a/src/mem/port.hh b/src/mem/port.hh index f9103865e..79fbf1fdb 100644 --- a/src/mem/port.hh +++ b/src/mem/port.hh @@ -92,11 +92,9 @@ class Port virtual ~Port() {}; // mey be better to use subclasses & RTTI? - /** Holds the ports status. Keeps track if it is blocked, or has - calculated a range change. */ + /** Holds the ports status. Currently just that a range recomputation needs + * to be done. */ enum Status { - Blocked, - Unblocked, RangeChange }; @@ -140,7 +138,7 @@ class Port wait. This shouldn't be valid for response paths (IO Devices). so it is set to panic if it isn't already defined. */ - virtual Packet *recvRetry() { panic("??"); } + virtual void recvRetry() { panic("??"); } /** Called by a peer port in order to determine the block size of the device connected to this port. It sometimes doesn't make sense for @@ -165,7 +163,7 @@ class Port port receive function. @return This function returns if the send was succesful in it's recieve. If it was a failure, then the port will wait for a recvRetry - at which point it can issue a successful sendTiming. This is used in + at which point it can possibly issue a successful sendTiming. This is used in case a cache has a higher priority request come in while waiting for the bus to arbitrate. */ @@ -194,7 +192,7 @@ class Port /** When a timing access doesn't return a success, some time later the Retry will be sent. */ - Packet *sendRetry() { return peer->recvRetry(); } + void sendRetry() { return peer->recvRetry(); } /** Called by the associated device if it wishes to find out the blocksize of the device on attached to the peer port.