Major changes to sinic device model. Rearrage read/write, better

interrupts.

dev/sinic.cc:
    - The prepareRead function sets all the variables in the register
    file that depend on various state bits that change on the fly.
    Includes RxDone, RxWait, TxDone, and TxWait
    - Use the new register information accessor functions to grab
    validity and size information for the read and write functions
    - read all registers directly from the register space by offset
    and size, not by actual name (less code)
    - The side effect of reading the interrupt status (clearing it) now
    happens outside the actual chunk of code where the value is loaded.
    - Add an iprRead function for when we may want speculative access
    to device registers through an ipr or special instruction.
    - When RxData or TxData are written, their busy flag is set to
    indicate that they have an outstanding transaction.
    - The RxHigh and TxLow interrupts are special, they only interrupt
    if the rxEmpty or txFull limits were hit
    - Move reset to the command register
    - Update more registers on reset, clear rxEmpty and txFull
    - Data dumps only happen if EthernetData trace flag set
    - When a DMA completes, kick the other engine if it was waiting
    - implement all of the new interrupts
    - serialize the new stuff
dev/sinic.hh:
    - Put all registers with their proper size and alignment into
    the regs struct so that we can copy multiple at a time.
    - Provide accessor functions for accessing the registers with
    different sizes.
    - Flags to track when the rx fifo hit empty and the tx fifo became
    full.  These flags are used to determine what to do when below
    the watermarks, and are reset when crossing the watermark.
    - the txDmaEvent should actually trigger the txDmaDone function
    - Add an iprRead function for when we may want speculative access
    to device registers through an ipr or special instruction.
    - The prepareRead function sets all the variables in the register
    file that depend on various state bits that change on the fly.
    - add rx_max_intr and dedicated (for dedicated thread) config params
dev/sinicreg.hh:
    Add some new registers: Command, RxMaxIntr, RxFifoSize, TxFifoSize,
    rename XxThreshold to XxFifoMark
    Move Reset to the Command register
    Add Thread to the Config register
    New interrupts, better names
    More info in RxDone and TxDone
    Easier access to information on each register (size, read, write, name)
python/m5/objects/Ethernet.py:
    Both sinic and nsgige have the dedicated thread
    Add a parameter to configure the maximum number for receive
    packets per interrupt

--HG--
extra : convert_revision : 407c5a993b6fb17326b4c623ee5d4b25fd69ac80

4 files changed, 328 insertions, 231 deletions

diff --git a/dev/sinic.cc b/dev/sinic.cc
index e3e1ffba0..1a1456e5f 100644
--- a/dev/sinic.cc
+++ b/dev/sinic.cc
@@ -316,9 +316,26 @@ Device::writeConfig(int offset, int size, const uint8_t *data)
     }
 }
 
+void
+Device::prepareRead()
+{
+    using namespace Regs;
+
+    // update rx registers
+    regs.RxDone = set_RxDone_Packets(regs.RxDone, rxFifo.packets());
+    regs.RxWait = regs.RxDone;
+
+    // update tx regsiters
+    regs.TxDone = set_TxDone_Packets(regs.TxDone, txFifo.packets());
+    regs.TxDone = set_TxDone_Full(regs.TxDone,
+                                  txFifo.avail() < regs.TxMaxCopy);
+    regs.TxDone = set_TxDone_Low(regs.TxDone,
+                                 txFifo.size() < regs.TxFifoMark);
+    regs.TxWait = regs.TxDone;
+}
+
 /**
- * This reads the device registers, which are detailed in the NS83820
- * spec sheet
+ * I/O read of device register
  */
 Fault
 Device::read(MemReqPtr &req, uint8_t *data)
@@ -328,118 +345,115 @@ Device::read(MemReqPtr &req, uint8_t *data)
     //The mask is to give you only the offset into the device register file
     Addr daddr = req->paddr & 0xfff;
 
-    if (Regs::regSize(daddr) == 0)
-        panic("invalid address: da=%#x pa=%#x va=%#x size=%d",
+    if (!regValid(daddr))
+        panic("invalid register: da=%#x pa=%#x va=%#x size=%d",
               daddr, req->paddr, req->vaddr, req->size);
 
-    if (req->size != Regs::regSize(daddr))
-        panic("invalid size for reg %s: da=%#x pa=%#x va=%#x size=%d",
-              Regs::regName(daddr), daddr, req->paddr, req->vaddr, req->size);
-
-    DPRINTF(EthernetPIO, "read reg=%s da=%#x pa=%#x va=%#x size=%d\n",
-            Regs::regName(daddr), daddr, req->paddr, req->vaddr, req->size);
-
-    uint32_t &reg32 = *(uint32_t *)data;
-    uint64_t &reg64 = *(uint64_t *)data;
+    const Regs::Info &info = regInfo(daddr);
+    if (!info.read)
+        panic("reading write only register: %s: da=%#x pa=%#x va=%#x size=%d",
+              info.name, daddr, req->paddr, req->vaddr, req->size);
 
-    switch (daddr) {
-      case Regs::Config:
-        reg32 = regs.Config;
-        break;
+    if (req->size != info.size)
+        panic("invalid size for reg %s: da=%#x pa=%#x va=%#x size=%d",
+              info.name, daddr, req->paddr, req->vaddr, req->size);
 
-      case Regs::RxMaxCopy:
-        reg32 = regs.RxMaxCopy;
-        break;
+    prepareRead();
 
-      case Regs::TxMaxCopy:
-        reg32 = regs.TxMaxCopy;
-        break;
+    uint64_t value = 0;
+    if (req->size == 4) {
+        uint32_t &reg = *(uint32_t *)data;
+        reg = regData32(daddr);
+        value = reg;
+    }
 
-      case Regs::RxThreshold:
-        reg32 = regs.RxThreshold;
-        break;
+    if (req->size == 8) {
+        uint64_t &reg = *(uint64_t *)data;
+        reg = regData64(daddr);
+        value = reg;
+    }
 
-      case Regs::TxThreshold:
-        reg32 = regs.TxThreshold;
-        break;
+    DPRINTF(EthernetPIO, "read reg=%s da=%#x pa=%#x va=%#x size=%d val=%#x\n",
+            info.name, daddr, req->paddr, req->vaddr, req->size, value);
 
-      case Regs::IntrStatus:
-        reg32 = regs.IntrStatus;
+    // reading the interrupt status register has the side effect of
+    // clearing it
+    if (daddr == Regs::IntrStatus)
         devIntrClear();
-        break;
 
-      case Regs::IntrMask:
-        reg32 = regs.IntrMask;
-        break;
+    return No_Fault;
+}
 
-      case Regs::RxData:
-        reg64 = regs.RxData;
-        break;
+/**
+ * IPR read of device register
+ */
+Fault
+Device::iprRead(Addr daddr, uint64_t &result)
+{
+    if (!regValid(daddr))
+        panic("invalid address: da=%#x", daddr);
 
-      case Regs::RxDone:
-      case Regs::RxWait:
-        reg64 = Regs::set_RxDone_FifoLen(regs.RxDone,
-                                         min(rxFifo.packets(), 255));
-        break;
+    const Regs::Info &info = regInfo(daddr);
+    if (!info.read)
+        panic("reading write only register %s: da=%#x", info.name, daddr);
 
-      case Regs::TxData:
-        reg64 = regs.TxData;
-        break;
+    DPRINTF(EthernetPIO, "read reg=%s da=%#x\n", info.name, daddr);
 
-      case Regs::TxDone:
-      case Regs::TxWait:
-        reg64 = Regs::set_TxDone_FifoLen(regs.TxDone,
-                                         min(txFifo.packets(), 255));
-        break;
+    prepareRead();
 
-      case Regs::HwAddr:
-        reg64 = params()->eaddr;
-        break;
+    if (info.size == 4)
+        result = regData32(daddr);
 
-      default:
-        panic("reading write only register %s: da=%#x pa=%#x va=%#x size=%d",
-              Regs::regName(daddr), daddr, req->paddr, req->vaddr, req->size);
-    }
+    if (info.size == 8)
+        result = regData64(daddr);
 
-    DPRINTF(EthernetPIO, "read reg=%s done val=%#x\n", Regs::regName(daddr),
-            Regs::regSize(daddr) == 4 ? reg32 : reg64);
+    DPRINTF(EthernetPIO, "IPR read reg=%s da=%#x val=%#x\n",
+            info.name, result);
 
     return No_Fault;
 }
 
+/**
+ * I/O write of device register
+ */
 Fault
 Device::write(MemReqPtr &req, const uint8_t *data)
 {
     assert(config.command & PCI_CMD_MSE);
+
+    //The mask is to give you only the offset into the device register file
     Addr daddr = req->paddr & 0xfff;
 
-    if (Regs::regSize(daddr) == 0)
+    if (!regValid(daddr))
         panic("invalid address: da=%#x pa=%#x va=%#x size=%d",
               daddr, req->paddr, req->vaddr, req->size);
 
-    if (req->size != Regs::regSize(daddr))
-        panic("invalid size: reg=%s da=%#x pa=%#x va=%#x size=%d",
-              Regs::regName(daddr), daddr, req->paddr, req->vaddr, req->size);
+    const Regs::Info &info = regInfo(daddr);
+    if (!info.write)
+        panic("writing read only register %s: da=%#x", info.name, daddr);
+
+    if (req->size != info.size)
+        panic("invalid size for reg %s: da=%#x pa=%#x va=%#x size=%d",
+              info.name, daddr, req->paddr, req->vaddr, req->size);
 
     uint32_t reg32 = *(uint32_t *)data;
     uint64_t reg64 = *(uint64_t *)data;
 
     DPRINTF(EthernetPIO, "write reg=%s val=%#x da=%#x pa=%#x va=%#x size=%d\n",
-            Regs::regName(daddr), Regs::regSize(daddr) == 4 ? reg32 : reg64,
-            daddr, req->paddr, req->vaddr, req->size);
-
+            info.name, info.size == 4 ? reg32 : reg64, daddr, req->paddr,
+            req->vaddr, req->size);
 
     switch (daddr) {
       case Regs::Config:
         changeConfig(reg32);
         break;
 
-      case Regs::RxThreshold:
-        regs.RxThreshold = reg32;
+      case Regs::Command:
+        command(reg32);
         break;
 
-      case Regs::TxThreshold:
-        regs.TxThreshold = reg32;
+      case Regs::IntrStatus:
+        devIntrClear(regs.IntrStatus & reg32);
         break;
 
       case Regs::IntrMask:
@@ -448,9 +462,10 @@ Device::write(MemReqPtr &req, const uint8_t *data)
 
       case Regs::RxData:
         if (rxState != rxIdle)
-            panic("receive machine busy with another request!");
+            panic("receive machine busy with another request! rxState=%s",
+                  RxStateStrings[rxState]);
 
-        regs.RxDone = 0;
+        regs.RxDone = Regs::RxDone_Busy;
         regs.RxData = reg64;
         if (rxEnable) {
             rxState = rxFifoBlock;
@@ -460,19 +475,16 @@ Device::write(MemReqPtr &req, const uint8_t *data)
 
       case Regs::TxData:
         if (txState != txIdle)
-            panic("transmit machine busy with another request!");
+            panic("transmit machine busy with another request! txState=%s",
+                  TxStateStrings[txState]);
 
-        regs.TxDone = 0;
+        regs.TxDone = Regs::TxDone_Busy;
         regs.TxData = reg64;
         if (txEnable) {
             txState = txFifoBlock;
             txKick();
         }
         break;
-
-      default:
-        panic("writing read only register %s: da=%#x pa=%#x va=%#x size=%d",
-              Regs::regName(daddr), daddr, req->paddr, req->vaddr, req->size);
     }
 
     return No_Fault;
@@ -490,9 +502,25 @@ Device::devIntrPost(uint32_t interrupts)
             "interrupt written to intStatus: intr=%#x status=%#x mask=%#x\n",
             interrupts, regs.IntrStatus, regs.IntrMask);
 
-    if ((regs.IntrStatus & regs.IntrMask)) {
+    interrupts = regs.IntrStatus & regs.IntrMask;
+
+    // Intr_RxHigh is special, we only signal it if we've emptied the fifo
+    // and then filled it above the high watermark
+    if (rxEmpty)
+        rxEmpty = false;
+    else
+        interrupts &= ~Regs::Intr_RxHigh;
+
+    // Intr_TxLow is special, we only signal it if we've filled up the fifo
+    // and then dropped below the low watermark
+    if (txFull)
+        txFull = false;
+    else
+        interrupts &= ~Regs::Intr_TxLow;
+
+    if (interrupts) {
         Tick when = curTick;
-        if ((regs.IntrStatus & regs.IntrMask & Regs::Intr_NoDelay) == 0)
+        if ((interrupts & Regs::Intr_NoDelay) == 0)
             when += intrDelay;
         cpuIntrPost(when);
     }
@@ -628,12 +656,6 @@ Device::changeConfig(uint32_t newconf)
 
     regs.Config = newconf;
 
-    if ((changed & Regs::Config_Reset)) {
-        assert(regs.Config & Regs::Config_Reset);
-        reset();
-        regs.Config &= ~Regs::Config_Reset;
-    }
-
     if ((changed & Regs::Config_IntEn)) {
         cpuIntrEnable = regs.Config & Regs::Config_IntEn;
         if (cpuIntrEnable) {
@@ -658,19 +680,39 @@ Device::changeConfig(uint32_t newconf)
 }
 
 void
+Device::command(uint32_t command)
+{
+    if (command & Regs::Command_Reset)
+        reset();
+}
+
+void
 Device::reset()
 {
     using namespace Regs;
+
     memset(&regs, 0, sizeof(regs));
+
+    regs.Config = 0;
+    if (params()->dedicated)
+        regs.Config |= Config_Thread;
+    regs.IntrMask = Intr_RxHigh | Intr_RxDMA | Intr_TxLow;
     regs.RxMaxCopy = params()->rx_max_copy;
     regs.TxMaxCopy = params()->tx_max_copy;
-    regs.IntrMask = Intr_TxFifo | Intr_RxFifo | Intr_RxData;
+    regs.RxMaxIntr = params()->rx_max_intr;
+    regs.RxFifoSize = params()->rx_fifo_size;
+    regs.TxFifoSize = params()->tx_fifo_size;
+    regs.RxFifoMark = params()->rx_fifo_threshold;
+    regs.TxFifoMark = params()->tx_fifo_threshold;
+    regs.HwAddr = params()->eaddr;
 
     rxState = rxIdle;
     txState = txIdle;
 
     rxFifo.clear();
     txFifo.clear();
+    rxEmpty = false;
+    txFull = false;
 }
 
 void
@@ -681,13 +723,18 @@ Device::rxDmaCopy()
     physmem->dma_write(rxDmaAddr, (uint8_t *)rxDmaData, rxDmaLen);
     DPRINTF(EthernetDMA, "rx dma write paddr=%#x len=%d\n",
             rxDmaAddr, rxDmaLen);
-    DDUMP(EthernetDMA, rxDmaData, rxDmaLen);
+    DDUMP(EthernetData, rxDmaData, rxDmaLen);
 }
 
 void
 Device::rxDmaDone()
 {
     rxDmaCopy();
+
+    // If the transmit state machine  has a pending DMA, let it go first
+    if (txState == txBeginCopy)
+        txKick();
+
     rxKick();
 }
 
@@ -707,6 +754,8 @@ Device::rxKick()
     switch (rxState) {
       case rxIdle:
         if (rxPioRequest) {
+            DPRINTF(EthernetPIO, "rxIdle: PIO waiting responding at %d\n",
+                    curTick + pioLatency);
             pioInterface->respond(rxPioRequest, curTick);
             rxPioRequest = 0;
         }
@@ -762,20 +811,20 @@ Device::rxKick()
         break;
 
       case rxBeginCopy:
+        if (dmaInterface && dmaInterface->busy())
+            goto exit;
+
         rxDmaAddr = plat->pciToDma(Regs::get_RxData_Addr(regs.RxData));
         rxDmaLen = min<int>(Regs::get_RxData_Len(regs.RxData), rxPktBytes);
         rxDmaData = rxPacketBufPtr;
+        rxState = rxCopy;
 
         if (dmaInterface) {
-            if (!dmaInterface->busy()) {
-                dmaInterface->doDMA(WriteInvalidate, rxDmaAddr, rxDmaLen,
-                                    curTick, &rxDmaEvent, true);
-                rxState = rxCopy;
-            }
+            dmaInterface->doDMA(WriteInvalidate, rxDmaAddr, rxDmaLen,
+                                curTick, &rxDmaEvent, true);
             goto exit;
         }
 
-        rxState = rxCopy;
         if (dmaWriteDelay != 0 || dmaWriteFactor != 0) {
             Tick factor = ((rxDmaLen + ULL(63)) >> ULL(6)) * dmaWriteFactor;
             Tick start = curTick + dmaWriteDelay + factor;
@@ -803,7 +852,7 @@ Device::rxKick()
         }
 
         regs.RxDone |= Regs::RxDone_Complete;
-        devIntrPost(Regs::Intr_RxData);
+        devIntrPost(Regs::Intr_RxDMA);
         rxState = rxIdle;
         break;
 
@@ -832,13 +881,18 @@ Device::txDmaCopy()
     physmem->dma_read((uint8_t *)txDmaData, txDmaAddr, txDmaLen);
     DPRINTF(EthernetDMA, "tx dma read paddr=%#x len=%d\n",
             txDmaAddr, txDmaLen);
-    DDUMP(EthernetDMA, txDmaData, txDmaLen);
+    DDUMP(EthernetData, txDmaData, txDmaLen);
 }
 
 void
 Device::txDmaDone()
 {
     txDmaCopy();
+
+    // If the receive state machine  has a pending DMA, let it go first
+    if (rxState == rxBeginCopy)
+        rxKick();
+
     txKick();
 }
 
@@ -850,6 +904,7 @@ Device::transmit()
         return;
     }
 
+    uint32_t interrupts;
     PacketPtr packet = txFifo.front();
     if (!interface->sendPacket(packet)) {
         DPRINTF(Ethernet, "Packet Transmit: failed txFifo available %d\n",
@@ -858,7 +913,6 @@ Device::transmit()
     }
 
     txFifo.pop();
-
 #if TRACING_ON
     if (DTRACE(Ethernet)) {
         IpPtr ip(packet);
@@ -873,17 +927,17 @@ Device::transmit()
     }
 #endif
 
-    DDUMP(Ethernet, packet->data, packet->length);
+    DDUMP(EthernetData, packet->data, packet->length);
     txBytes += packet->length;
     txPackets++;
 
     DPRINTF(Ethernet, "Packet Transmit: successful txFifo Available %d\n",
             txFifo.avail());
 
-    if (txFifo.size() <= params()->tx_fifo_threshold)
-        devIntrPost(Regs::Intr_TxFifo);
-
-    devIntrPost(Regs::Intr_TxDone);
+    interrupts = Regs::Intr_TxPacket;
+    if (txFifo.size() < regs.TxFifoMark)
+        interrupts |= Regs::Intr_TxLow;
+    devIntrPost(interrupts);
 
   reschedule:
    if (!txFifo.empty() && !txEvent.scheduled()) {
@@ -908,6 +962,8 @@ Device::txKick()
     switch (txState) {
       case txIdle:
         if (txPioRequest) {
+            DPRINTF(EthernetPIO, "txIdle: PIO waiting responding at %d\n",
+                    curTick + pioLatency);
             pioInterface->respond(txPioRequest, curTick + pioLatency);
             txPioRequest = 0;
         }
@@ -930,21 +986,20 @@ Device::txKick()
         break;
 
       case txBeginCopy:
+        if (dmaInterface && dmaInterface->busy())
+            goto exit;
+
         txDmaAddr = plat->pciToDma(Regs::get_TxData_Addr(regs.TxData));
         txDmaLen = Regs::get_TxData_Len(regs.TxData);
         txDmaData = txPacketBufPtr;
+        txState = txCopy;
 
         if (dmaInterface) {
-            if (!dmaInterface->busy()) {
-                dmaInterface->doDMA(Read, txDmaAddr, txDmaLen,
-                                    curTick, &txDmaEvent, true);
-                txState = txCopy;
-            }
-
+            dmaInterface->doDMA(Read, txDmaAddr, txDmaLen,
+                                curTick, &txDmaEvent, true);
             goto exit;
         }
 
-        txState = txCopy;
         if (dmaReadDelay != 0 || dmaReadFactor != 0) {
             Tick factor = ((txDmaLen + ULL(63)) >> ULL(6)) * dmaReadFactor;
             Tick start = curTick + dmaReadDelay + factor;
@@ -988,12 +1043,16 @@ Device::txKick()
                 }
             }
             txFifo.push(txPacket);
+            if (txFifo.avail() < regs.TxMaxCopy) {
+                devIntrPost(Regs::Intr_TxFull);
+                txFull = true;
+            }
             txPacket = 0;
             transmit();
         }
 
         regs.TxDone = txDmaLen | Regs::TxDone_Complete;
-        devIntrPost(Regs::Intr_TxData);
+        devIntrPost(Regs::Intr_TxDMA);
         txState = txIdle;
         break;
 
@@ -1094,8 +1153,8 @@ Device::recvPacket(PacketPtr packet)
         return true;
     }
 
-    if (rxFifo.size() >= params()->rx_fifo_threshold)
-        devIntrPost(Regs::Intr_RxFifo);
+    if (rxFifo.size() >= regs.RxFifoMark)
+        devIntrPost(Regs::Intr_RxHigh);
 
     if (!rxFifo.push(packet)) {
         DPRINTF(Ethernet,
@@ -1103,7 +1162,7 @@ Device::recvPacket(PacketPtr packet)
         return false;
     }
 
-    devIntrPost(Regs::Intr_RxDone);
+    devIntrPost(Regs::Intr_RxPacket);
     rxKick();
     return true;
 }
@@ -1161,22 +1220,23 @@ Device::serialize(ostream &os)
     // Serialize the PciDev base class
     Base::serialize(os);
 
-    if (rxDmaEvent.scheduled())
-        rxDmaCopy();
+    if (rxState == rxCopy)
+        panic("can't serialize with an in flight dma request rxState=%s",
+              RxStateStrings[rxState]);
 
-    if (txDmaEvent.scheduled())
-        txDmaCopy();
+    if (txState == txCopy)
+        panic("can't serialize with an in flight dma request txState=%s",
+              TxStateStrings[txState]);
 
     /*
      * Serialize the device registers
      */
     SERIALIZE_SCALAR(regs.Config);
-    SERIALIZE_SCALAR(regs.RxMaxCopy);
-    SERIALIZE_SCALAR(regs.TxMaxCopy);
-    SERIALIZE_SCALAR(regs.RxThreshold);
-    SERIALIZE_SCALAR(regs.TxThreshold);
     SERIALIZE_SCALAR(regs.IntrStatus);
     SERIALIZE_SCALAR(regs.IntrMask);
+    SERIALIZE_SCALAR(regs.RxMaxCopy);
+    SERIALIZE_SCALAR(regs.TxMaxCopy);
+    SERIALIZE_SCALAR(regs.RxMaxIntr);
     SERIALIZE_SCALAR(regs.RxData);
     SERIALIZE_SCALAR(regs.RxDone);
     SERIALIZE_SCALAR(regs.TxData);
@@ -1187,6 +1247,7 @@ Device::serialize(ostream &os)
      */
     int rxState = this->rxState;
     SERIALIZE_SCALAR(rxState);
+    SERIALIZE_SCALAR(rxEmpty);
     rxFifo.serialize("rxFifo", os);
     bool rxPacketExists = rxPacket;
     SERIALIZE_SCALAR(rxPacketExists);
@@ -1203,6 +1264,7 @@ Device::serialize(ostream &os)
      */
     int txState = this->txState;
     SERIALIZE_SCALAR(txState);
+    SERIALIZE_SCALAR(txFull);
     txFifo.serialize("txFifo", os);
     bool txPacketExists = txPacket;
     SERIALIZE_SCALAR(txPacketExists);
@@ -1231,12 +1293,11 @@ Device::unserialize(Checkpoint *cp, const std::string &section)
      * Unserialize the device registers
      */
     UNSERIALIZE_SCALAR(regs.Config);
-    UNSERIALIZE_SCALAR(regs.RxMaxCopy);
-    UNSERIALIZE_SCALAR(regs.TxMaxCopy);
-    UNSERIALIZE_SCALAR(regs.RxThreshold);
-    UNSERIALIZE_SCALAR(regs.TxThreshold);
     UNSERIALIZE_SCALAR(regs.IntrStatus);
     UNSERIALIZE_SCALAR(regs.IntrMask);
+    UNSERIALIZE_SCALAR(regs.RxMaxCopy);
+    UNSERIALIZE_SCALAR(regs.TxMaxCopy);
+    UNSERIALIZE_SCALAR(regs.RxMaxIntr);
     UNSERIALIZE_SCALAR(regs.RxData);
     UNSERIALIZE_SCALAR(regs.RxDone);
     UNSERIALIZE_SCALAR(regs.TxData);
@@ -1247,6 +1308,7 @@ Device::unserialize(Checkpoint *cp, const std::string &section)
      */
     int rxState;
     UNSERIALIZE_SCALAR(rxState);
+    UNSERIALIZE_SCALAR(rxEmpty);
     this->rxState = (RxState) rxState;
     rxFifo.unserialize("rxFifo", cp, section);
     bool rxPacketExists;
@@ -1267,6 +1329,7 @@ Device::unserialize(Checkpoint *cp, const std::string &section)
      */
     int txState;
     UNSERIALIZE_SCALAR(txState);
+    UNSERIALIZE_SCALAR(txFull);
     this->txState = (TxState) txState;
     txFifo.unserialize("txFifo", cp, section);
     bool txPacketExists;
@@ -1308,15 +1371,19 @@ Device::cacheAccess(MemReqPtr &req)
     Tick when = curTick + pioLatency;
 
     switch (daddr) {
-      case Regs::RxDone:
+      case Regs::RxWait:
         if (rxState != rxIdle) {
+            DPRINTF(EthernetPIO, "rxState=%s (not idle)... waiting\n",
+                    TxStateStrings[txState]);
             rxPioRequest = req;
             when = 0;
         }
         break;
 
-      case Regs::TxDone:
+      case Regs::TxWait:
         if (txState != txIdle) {
+            DPRINTF(EthernetPIO, "txState=%s (not idle)... waiting\n",
+                    TxStateStrings[txState]);
             txPioRequest = req;
             when = 0;
         }
@@ -1385,6 +1452,7 @@ BEGIN_DECLARE_SIM_OBJECT_PARAMS(Device)
     Param<Tick> tx_delay;
     Param<uint32_t> rx_max_copy;
     Param<uint32_t> tx_max_copy;
+    Param<uint32_t> rx_max_intr;
     Param<uint32_t> rx_fifo_size;
     Param<uint32_t> tx_fifo_size;
     Param<uint32_t> rx_fifo_threshold;
@@ -1392,6 +1460,7 @@ BEGIN_DECLARE_SIM_OBJECT_PARAMS(Device)
 
     Param<bool> rx_filter;
     Param<string> hardware_address;
+    Param<bool> dedicated;
 
 END_DECLARE_SIM_OBJECT_PARAMS(Device)
 
@@ -1424,13 +1493,15 @@ BEGIN_INIT_SIM_OBJECT_PARAMS(Device)
     INIT_PARAM(tx_delay, "Transmit Delay"),
     INIT_PARAM(rx_max_copy, "rx max copy"),
     INIT_PARAM(tx_max_copy, "rx max copy"),
+    INIT_PARAM(rx_max_intr, "rx max intr"),
     INIT_PARAM(rx_fifo_size, "max size in bytes of rxFifo"),
     INIT_PARAM(tx_fifo_size, "max size in bytes of txFifo"),
     INIT_PARAM(rx_fifo_threshold, "max size in bytes of rxFifo"),
     INIT_PARAM(tx_fifo_threshold, "max size in bytes of txFifo"),
 
     INIT_PARAM(rx_filter, "Enable Receive Filter"),
-    INIT_PARAM(hardware_address, "Ethernet Hardware Address")
+    INIT_PARAM(hardware_address, "Ethernet Hardware Address"),
+    INIT_PARAM(dedicated, "dedicate a kernel thread to the driver")
 
 END_INIT_SIM_OBJECT_PARAMS(Device)
 
@@ -1442,6 +1513,7 @@ CREATE_SIM_OBJECT(Device)
     params->name = getInstanceName();
 
     params->clock = clock;
+
     params->mmu = mmu;
     params->physmem = physmem;
     params->configSpace = configspace;
@@ -1466,6 +1538,7 @@ CREATE_SIM_OBJECT(Device)
     params->rx_delay = rx_delay;
     params->rx_max_copy = rx_max_copy;
     params->tx_max_copy = tx_max_copy;
+    params->rx_max_intr = rx_max_intr;
     params->rx_fifo_size = rx_fifo_size;
     params->tx_fifo_size = tx_fifo_size;
     params->rx_fifo_threshold = rx_fifo_threshold;
@@ -1473,6 +1546,7 @@ CREATE_SIM_OBJECT(Device)
 
     params->rx_filter = rx_filter;
     params->eaddr = hardware_address;
+    params->dedicated = dedicated;
 
     return new Device(params);
 }
diff --git a/dev/sinic.hh b/dev/sinic.hh
index 924c6eeeb..e01015061 100644
--- a/dev/sinic.hh
+++ b/dev/sinic.hh
@@ -115,19 +115,31 @@ class Device : public Base
 
     /** device register file */
     struct {
-        uint32_t Config;
-        uint32_t RxMaxCopy;
-        uint32_t TxMaxCopy;
-        uint32_t RxThreshold;
-        uint32_t TxThreshold;
-        uint32_t IntrStatus;
-        uint32_t IntrMask;
-        uint64_t RxData;
-        uint64_t RxDone;
-        uint64_t TxData;
-        uint64_t TxDone;
+        uint32_t Config;       // 0x00
+        uint32_t Command;      // 0x04
+        uint32_t IntrStatus;   // 0x08
+        uint32_t IntrMask;     // 0x0c
+        uint32_t RxMaxCopy;    // 0x10
+        uint32_t TxMaxCopy;    // 0x14
+        uint32_t RxMaxIntr;    // 0x18
+        uint32_t Reserved0;    // 0x1c
+        uint32_t RxFifoSize;   // 0x20
+        uint32_t TxFifoSize;   // 0x24
+        uint32_t RxFifoMark;   // 0x28
+        uint32_t TxFifoMark;   // 0x2c
+        uint64_t RxData;       // 0x30
+        uint64_t RxDone;       // 0x38
+        uint64_t RxWait;       // 0x40
+        uint64_t TxData;       // 0x48
+        uint64_t TxDone;       // 0x50
+        uint64_t TxWait;       // 0x58
+        uint64_t HwAddr;       // 0x60
     } regs;
 
+    uint8_t  &regData8(Addr daddr) { return *((uint8_t *)&regs + daddr); }
+    uint32_t &regData32(Addr daddr) { return *(uint32_t *)&regData8(daddr); }
+    uint64_t &regData64(Addr daddr) { return *(uint64_t *)&regData8(daddr); }
+
   private:
     Addr addr;
     static const Addr size = Regs::Size;
@@ -135,6 +147,7 @@ class Device : public Base
   protected:
     RxState rxState;
     PacketFifo rxFifo;
+    bool rxEmpty;
     PacketPtr rxPacket;
     uint8_t *rxPacketBufPtr;
     int rxPktBytes;
@@ -145,6 +158,7 @@ class Device : public Base
 
     TxState txState;
     PacketFifo txFifo;
+    bool txFull;
     PacketPtr txPacket;
     uint8_t *txPacketBufPtr;
     int txPktBytes;
@@ -191,6 +205,7 @@ class Device : public Base
  * device configuration
  */
     void changeConfig(uint32_t newconfig);
+    void command(uint32_t command);
 
 /**
  * device ethernet interface
@@ -212,7 +227,7 @@ class Device : public Base
     void txDmaCopy();
     void txDmaDone();
     friend class EventWrapper<Device, &Device::txDmaDone>;
-    EventWrapper<Device, &Device::rxDmaDone> txDmaEvent;
+    EventWrapper<Device, &Device::txDmaDone> txDmaEvent;
 
     Tick dmaReadDelay;
     Tick dmaReadFactor;
@@ -244,6 +259,8 @@ class Device : public Base
  * Memory Interface
  */
   public:
+    void prepareRead();
+    Fault iprRead(Addr daddr, uint64_t &result);
     virtual Fault read(MemReqPtr &req, uint8_t *data);
     virtual Fault write(MemReqPtr &req, const uint8_t *data);
     Tick cacheAccess(MemReqPtr &req);
@@ -308,6 +325,7 @@ class Device : public Base
         Net::EthAddr eaddr;
         uint32_t rx_max_copy;
         uint32_t tx_max_copy;
+        uint32_t rx_max_intr;
         uint32_t rx_fifo_size;
         uint32_t tx_fifo_size;
         uint32_t rx_fifo_threshold;
@@ -317,6 +335,7 @@ class Device : public Base
         Tick dma_write_delay;
         Tick dma_write_factor;
         bool dma_no_allocate;
+        bool dedicated;
     };
 
   protected:
diff --git a/dev/sinicreg.hh b/dev/sinicreg.hh
index 78b175f71..12f545255 100644
--- a/dev/sinicreg.hh
+++ b/dev/sinicreg.hh
@@ -57,23 +57,28 @@ namespace Regs {
 
 // Registers
 __SINIC_REG32(Config,      0x00); // 32: configuration register
-__SINIC_REG32(RxMaxCopy,   0x04); // 32: max rx copy
-__SINIC_REG32(TxMaxCopy,   0x08); // 32: max tx copy
-__SINIC_REG32(RxThreshold, 0x0c); // 32: receive fifo threshold
-__SINIC_REG32(TxThreshold, 0x10); // 32: transmit fifo threshold
-__SINIC_REG32(IntrStatus,  0x14); // 32: interrupt status
-__SINIC_REG32(IntrMask,    0x18); // 32: interrupt mask
-__SINIC_REG32(RxData,      0x20); // 64: receive data
-__SINIC_REG32(RxDone,      0x28); // 64: receive done
-__SINIC_REG32(RxWait,      0x30); // 64: receive done (busy wait)
-__SINIC_REG32(TxData,      0x38); // 64: transmit data
-__SINIC_REG32(TxDone,      0x40); // 64: transmit done
-__SINIC_REG32(TxWait,      0x48); // 64: transmit done (busy wait)
-__SINIC_REG32(HwAddr,      0x50); // 64: mac address
-__SINIC_REG32(Size,        0x58);
+__SINIC_REG32(Command,     0x04); // 32: command register
+__SINIC_REG32(IntrStatus,  0x08); // 32: interrupt status
+__SINIC_REG32(IntrMask,    0x0c); // 32: interrupt mask
+__SINIC_REG32(RxMaxCopy,   0x10); // 32: max bytes per rx copy
+__SINIC_REG32(TxMaxCopy,   0x14); // 32: max bytes per tx copy
+__SINIC_REG32(RxMaxIntr,   0x18); // 32: max receives per interrupt
+__SINIC_REG32(Reserved0,   0x1c); // 32: reserved
+__SINIC_REG32(RxFifoSize,  0x20); // 32: rx fifo capacity in bytes
+__SINIC_REG32(TxFifoSize,  0x24); // 32: tx fifo capacity in bytes
+__SINIC_REG32(RxFifoMark,  0x28); // 32: rx fifo high watermark
+__SINIC_REG32(TxFifoMark,  0x2c); // 32: tx fifo low watermark
+__SINIC_REG32(RxData,      0x30); // 64: receive data
+__SINIC_REG32(RxDone,      0x38); // 64: receive done
+__SINIC_REG32(RxWait,      0x40); // 64: receive done (busy wait)
+__SINIC_REG32(TxData,      0x48); // 64: transmit data
+__SINIC_REG32(TxDone,      0x50); // 64: transmit done
+__SINIC_REG32(TxWait,      0x58); // 64: transmit done (busy wait)
+__SINIC_REG32(HwAddr,      0x60); // 64: mac address
+__SINIC_REG32(Size,        0x68); // register addres space size
 
 // Config register bits
-__SINIC_VAL32(Config_Reset,  31, 1); // reset chip
+__SINIC_VAL32(Config_Thread,  8, 1); // enable receive filter
 __SINIC_VAL32(Config_Filter,  7, 1); // enable receive filter
 __SINIC_VAL32(Config_Vlan,    6, 1); // enable vlan tagging
 __SINIC_VAL32(Config_Virtual, 5, 1); // enable virtual addressing
@@ -83,105 +88,103 @@ __SINIC_VAL32(Config_IntEn,   2, 1); // enable interrupts
 __SINIC_VAL32(Config_TxEn,    1, 1); // enable transmit
 __SINIC_VAL32(Config_RxEn,    0, 1); // enable receive
 
+// Command register bits
+__SINIC_VAL32(Command_Reset, 0, 1); // reset chip
+
 // Interrupt register bits
-__SINIC_VAL32(Intr_TxFifo, 5, 1);  // Fifo oflow/uflow/threshold
-__SINIC_VAL32(Intr_TxData, 4, 1);  // DMA Completed w/ interrupt
-__SINIC_VAL32(Intr_TxDone, 3, 1);  // Packet transmitted
-__SINIC_VAL32(Intr_RxFifo, 2, 1); // Fifo oflow/uflow/threshold
-__SINIC_VAL32(Intr_RxData, 1, 1); // DMA Completed w/ interrupt
-__SINIC_VAL32(Intr_RxDone, 0, 1); // Packet received
-__SINIC_REG32(Intr_All,     0x3f);
-__SINIC_REG32(Intr_NoDelay, 0x24);
-__SINIC_REG32(Intr_Res,    ~0x3f);
+__SINIC_VAL32(Intr_TxLow,     7, 1); // tx fifo dropped below watermark
+__SINIC_VAL32(Intr_TxFull,    6, 1); // tx fifo full
+__SINIC_VAL32(Intr_TxDMA,     5, 1); // tx dma completed w/ interrupt
+__SINIC_VAL32(Intr_TxPacket,  4, 1); // packet transmitted
+__SINIC_VAL32(Intr_RxHigh,    3, 1); // rx fifo above high watermark
+__SINIC_VAL32(Intr_RxEmpty,   2, 1); // rx fifo empty
+__SINIC_VAL32(Intr_RxDMA,     1, 1); // rx dma completed w/ interrupt
+__SINIC_VAL32(Intr_RxPacket,  0, 1); // packet received
+__SINIC_REG32(Intr_All,       0xff); // all valid interrupts
+__SINIC_REG32(Intr_NoDelay,   0xcc); // interrupts that shouldn't be coalesced
+__SINIC_REG32(Intr_Res,      ~0xff); // reserved interrupt bits
 
 // RX Data Description
 __SINIC_VAL64(RxData_Len, 40, 20); // 0 - 1M
 __SINIC_VAL64(RxData_Addr, 0, 40); // Address 1TB
 
 // TX Data Description
-__SINIC_VAL64(TxData_More,     63,  1);
-__SINIC_VAL64(TxData_Checksum, 62,  1);
+__SINIC_VAL64(TxData_More,     63,  1); // Packet not complete (will dma more)
+__SINIC_VAL64(TxData_Checksum, 62,  1); // do checksum
 __SINIC_VAL64(TxData_Len,      40, 20); // 0 - 1M
 __SINIC_VAL64(TxData_Addr,      0, 40); // Address 1TB
 
 // RX Done/Busy Information
-__SINIC_VAL64(RxDone_Complete,  63,  1);
-__SINIC_VAL64(RxDone_IpPacket,  45,  1);
-__SINIC_VAL64(RxDone_TcpPacket, 44,  1);
-__SINIC_VAL64(RxDone_UdpPacket, 43,  1);
-__SINIC_VAL64(RxDone_IpError,   42,  1);
-__SINIC_VAL64(RxDone_TcpError,  41,  1);
-__SINIC_VAL64(RxDone_UdpError,  40,  1);
-__SINIC_VAL64(RxDone_More,      32,  1);
-__SINIC_VAL64(RxDone_FifoLen,   20,  8); // up to 255 packets
+__SINIC_VAL64(RxDone_Packets,   32, 16); // number of packets in rx fifo
+__SINIC_VAL64(RxDone_Busy,      31,  1); // receive dma busy copying
+__SINIC_VAL64(RxDone_Complete,  30,  1); // valid data (packet complete)
+__SINIC_VAL64(RxDone_More,      29,  1); // Packet has more data (dma again)
+__SINIC_VAL64(RxDone_TcpError,  25,  1); // TCP packet error (bad checksum)
+__SINIC_VAL64(RxDone_UdpError,  24,  1); // UDP packet error (bad checksum)
+__SINIC_VAL64(RxDone_IpError,   23,  1); // IP packet error (bad checksum)
+__SINIC_VAL64(RxDone_TcpPacket, 22,  1); // this is a TCP packet
+__SINIC_VAL64(RxDone_UdpPacket, 21,  1); // this is a UDP packet
+__SINIC_VAL64(RxDone_IpPacket,  20,  1); // this is an IP packet
 __SINIC_VAL64(RxDone_CopyLen,    0, 20); // up to 256k
 
 // TX Done/Busy Information
-__SINIC_VAL64(TxDone_Complete, 63,  1);
-__SINIC_VAL64(TxDone_FifoLen,  20,  8); // up to 255 packets
-__SINIC_VAL64(TxDone_CopyLen,   0, 20); // up to 256k
-
-inline int
-regSize(int offset)
+__SINIC_VAL64(TxDone_Packets,   32, 16); // number of packets in tx fifo
+__SINIC_VAL64(TxDone_Busy,      31,  1); // transmit dma busy copying
+__SINIC_VAL64(TxDone_Complete,  30,  1); // valid data (packet complete)
+__SINIC_VAL64(TxDone_Full,      29,  1); // tx fifo is full
+__SINIC_VAL64(TxDone_Low,       28,  1); // tx fifo is below the watermark
+__SINIC_VAL64(TxDone_CopyLen,    0, 20); // up to 256k
+
+struct Info
 {
-    static const char sizes[] = {
-        4,
-        4,
-        4,
-        4,
-        4,
-        4,
-        4,
-        0,
-        8, 0,
-        8, 0,
-        8, 0,
-        8, 0,
-        8, 0,
-        8, 0,
-        8, 0
-    };
+    uint8_t size;
+    bool read;
+    bool write;
+    const char *name;
+};
 
-    if (offset & 0x3)
-        return 0;
+/* namespace Regs */ }
 
-    if (offset >= Size)
-        return 0;
+inline const Regs::Info&
+regInfo(Addr daddr)
+{
+    static Regs::Info info [] = {
+        { 4, true,  true,  "Config"     },
+        { 4, false, true,  "Command"    },
+        { 4, true,  true,  "IntrStatus" },
+        { 4, true,  true,  "IntrMask"   },
+        { 4, true,  false, "RxMaxCopy"  },
+        { 4, true,  false, "TxMaxCopy"  },
+        { 4, true,  false, "RxMaxIntr"  },
+        { 0, false, false, "invalid"    },
+        { 4, true,  false, "RxFifoSize" },
+        { 4, true,  false, "TxFifoSize" },
+        { 4, true,  false, "RxFifoMark" },
+        { 4, true,  false, "TxFifoMark" },
+        { 8, true,  true,  "RxData"     }, { 0, false, false, "invalid" },
+        { 8, true,  false, "RxDone"     }, { 0, false, false, "invalid" },
+        { 8, true,  false, "RxWait"     }, { 0, false, false, "invalid" },
+        { 8, true,  true,  "TxData"     }, { 0, false, false, "invalid" },
+        { 8, true,  false, "TxDone"     }, { 0, false, false, "invalid" },
+        { 8, true,  false, "TxWait"     }, { 0, false, false, "invalid" },
+        { 8, true,  false, "HwAddr"     }, { 0, false, false, "invalid" }
+    };
 
-    return sizes[offset / 4];
+    return info[daddr / 4];
 }
 
-inline const char *
-regName(int offset)
+inline bool
+regValid(Addr daddr)
 {
-    static const char *names[] = {
-        "Config",
-        "RxMaxCopy",
-        "TxMaxCopy",
-        "RxThreshold",
-        "TxThreshold",
-        "IntrStatus",
-        "IntrMask",
-        "invalid",
-        "RxData", "invalid",
-        "RxDone", "invalid",
-        "RxWait", "invalid",
-        "TxData", "invalid",
-        "TxDone", "invalid",
-        "TxWait", "invalid",
-        "HwAddr", "invalid"
-    };
+    if (daddr > Regs::Size)
+        return false;
 
-    if (offset & 0x3)
-        return "invalid";
+    if (regInfo(daddr).size == 0)
+        return false;
 
-    if (offset >= Size)
-        return "invalid";
-
-    return names[offset / 4];
+    return true;
 }
 
-/* namespace Regs */ }
 /* namespace Sinic */ }
 
 #endif // __DEV_SINICREG_HH__
diff --git a/python/m5/objects/Ethernet.py b/python/m5/objects/Ethernet.py
index c2f818325..a97e58bda 100644
--- a/python/m5/objects/Ethernet.py
+++ b/python/m5/objects/Ethernet.py
@@ -83,6 +83,7 @@ class EtherDevBase(PciDevice):
 
     rx_filter = Param.Bool(True, "Enable Receive Filter")
     intr_delay = Param.Latency('10us', "Interrupt Propagation Delay")
+    dedicated = Param.Bool(False, "dedicate a kernel thread to the driver")
 
 class NSGigE(EtherDevBase):
     type = 'NSGigE'
@@ -90,7 +91,6 @@ class NSGigE(EtherDevBase):
     dma_data_free = Param.Bool(False, "DMA of Data is free")
     dma_desc_free = Param.Bool(False, "DMA of Descriptors is free")
 
-    dedicated = Param.Bool(False, "dedicate a kernel thread to the driver")
 
 class NSGigEInt(EtherInt):
     type = 'NSGigEInt'
@@ -101,6 +101,7 @@ class Sinic(EtherDevBase):
 
     rx_max_copy = Param.MemorySize('1514B', "rx max copy")
     tx_max_copy = Param.MemorySize('16kB', "tx max copy")
+    rx_max_intr = Param.UInt32(10, "max rx packets per interrupt")
     rx_fifo_threshold = Param.MemorySize('48kB', "rx fifo high threshold")
     tx_fifo_threshold = Param.MemorySize('16kB', "tx fifo low threshold")