diff --git a/src/main/scala/groundtest/Regression.scala b/src/main/scala/groundtest/Regression.scala
index 65289a23..ff8010a6 100644
--- a/src/main/scala/groundtest/Regression.scala
+++ b/src/main/scala/groundtest/Regression.scala
@@ -5,8 +5,8 @@ import uncore.tilelink._
 import uncore.constants._
 import uncore.agents._
 import uncore.util._
-import junctions.{ParameterizedBundle, HasAddrMapParameters}
-import util.Timer
+import junctions.HasAddrMapParameters
+import util.{ParameterizedBundle, Timer}
 import rocket.HellaCacheIO
 import cde.{Parameters, Field}
 
diff --git a/src/main/scala/groundtest/Tile.scala b/src/main/scala/groundtest/Tile.scala
index b0b232d9..7745d333 100644
--- a/src/main/scala/groundtest/Tile.scala
+++ b/src/main/scala/groundtest/Tile.scala
@@ -3,9 +3,10 @@ package groundtest
 import Chisel._
 import rocket._
 import uncore.tilelink._
-import junctions._
 import scala.util.Random
 import scala.collection.mutable.ListBuffer
+import junctions.HasAddrMapParameters
+import util.ParameterizedBundle
 import cde.{Parameters, Field}
 
 case object BuildGroundTest extends Field[Parameters => GroundTest]
diff --git a/src/main/scala/junctions/hasti.scala b/src/main/scala/junctions/hasti.scala
index 0f25d19b..6063729e 100644
--- a/src/main/scala/junctions/hasti.scala
+++ b/src/main/scala/junctions/hasti.scala
@@ -3,6 +3,7 @@ package junctions
 import Chisel._
 import cde.{Parameters, Field}
 import unittest.UnitTest
+import util.ParameterizedBundle
 
 object HastiConstants
 {
diff --git a/src/main/scala/junctions/memserdes.scala b/src/main/scala/junctions/memserdes.scala
index 14cac184..e34205ac 100644
--- a/src/main/scala/junctions/memserdes.scala
+++ b/src/main/scala/junctions/memserdes.scala
@@ -3,6 +3,7 @@
 package junctions
 import Chisel._
 import scala.math._
+import util.{HellaQueue, ParameterizedBundle}
 import cde.{Parameters, Field}
 
 case object MIFAddrBits extends Field[Int]
diff --git a/src/main/scala/junctions/nasti.scala b/src/main/scala/junctions/nasti.scala
index 43803153..48cd4fb5 100644
--- a/src/main/scala/junctions/nasti.scala
+++ b/src/main/scala/junctions/nasti.scala
@@ -4,6 +4,7 @@ package junctions
 import Chisel._
 import scala.math.max
 import scala.collection.mutable.ArraySeq
+import util.{ParameterizedBundle, HellaPeekingArbiter}
 import cde.{Parameters, Field}
 
 case object NastiKey extends Field[NastiParameters]
@@ -449,7 +450,7 @@ class NastiRouter(nSlaves: Int, routeSel: UInt => UInt)(implicit p: Parameters)
   io.master.w.ready := w_ready || err_slave.io.w.ready
 
   val b_arb = Module(new RRArbiter(new NastiWriteResponseChannel, nSlaves + 1))
-  val r_arb = Module(new JunctionsPeekingArbiter(
+  val r_arb = Module(new HellaPeekingArbiter(
     new NastiReadDataChannel, nSlaves + 1,
     // we can unlock if it's the last beat
     (r: NastiReadDataChannel) => r.last))
diff --git a/src/main/scala/junctions/util.scala b/src/main/scala/junctions/util.scala
deleted file mode 100644
index 109433f1..00000000
--- a/src/main/scala/junctions/util.scala
+++ /dev/null
@@ -1,246 +0,0 @@
-/// See LICENSE for license details.
-package junctions
-import Chisel._
-import cde.Parameters
-
-class ParameterizedBundle(implicit p: Parameters) extends Bundle {
-  override def cloneType = {
-    try {
-      this.getClass.getConstructors.head.newInstance(p).asInstanceOf[this.type]
-    } catch {
-      case e: java.lang.IllegalArgumentException =>
-        throwException("Unable to use ParamaterizedBundle.cloneType on " +
-                       this.getClass + ", probably because " + this.getClass +
-                       "() takes more than one argument.  Consider overriding " +
-                       "cloneType() on " + this.getClass, e)
-    }
-  }
-}
-
-class HellaFlowQueue[T <: Data](val entries: Int)(data: => T) extends Module {
-  val io = new QueueIO(data, entries)
-  require(entries > 1)
-
-  val do_flow = Wire(Bool())
-  val do_enq = io.enq.fire() && !do_flow
-  val do_deq = io.deq.fire() && !do_flow
-
-  val maybe_full = Reg(init=Bool(false))
-  val enq_ptr = Counter(do_enq, entries)._1
-  val (deq_ptr, deq_done) = Counter(do_deq, entries)
-  when (do_enq =/= do_deq) { maybe_full := do_enq }
-
-  val ptr_match = enq_ptr === deq_ptr
-  val empty = ptr_match && !maybe_full
-  val full = ptr_match && maybe_full
-  val atLeastTwo = full || enq_ptr - deq_ptr >= UInt(2)
-  do_flow := empty && io.deq.ready
-
-  val ram = SeqMem(entries, data)
-  when (do_enq) { ram.write(enq_ptr, io.enq.bits) }
-
-  val ren = io.deq.ready && (atLeastTwo || !io.deq.valid && !empty)
-  val raddr = Mux(io.deq.valid, Mux(deq_done, UInt(0), deq_ptr + UInt(1)), deq_ptr)
-  val ram_out_valid = Reg(next = ren)
-
-  io.deq.valid := Mux(empty, io.enq.valid, ram_out_valid)
-  io.enq.ready := !full
-  io.deq.bits := Mux(empty, io.enq.bits, ram.read(raddr, ren))
-}
-
-class HellaQueue[T <: Data](val entries: Int)(data: => T) extends Module {
-  val io = new QueueIO(data, entries)
-
-  val fq = Module(new HellaFlowQueue(entries)(data))
-  fq.io.enq <> io.enq
-  io.deq <> Queue(fq.io.deq, 1, pipe = true)
-}
-
-object HellaQueue {
-  def apply[T <: Data](enq: DecoupledIO[T], entries: Int) = {
-    val q = Module((new HellaQueue(entries)) { enq.bits })
-    q.io.enq.valid := enq.valid // not using <> so that override is allowed
-    q.io.enq.bits := enq.bits
-    enq.ready := q.io.enq.ready
-    q.io.deq
-  }
-}
-
-/** A generalized locking RR arbiter that addresses the limitations of the
- *  version in the Chisel standard library */
-abstract class JunctionsAbstractLockingArbiter[T <: Data](typ: T, arbN: Int, rr: Boolean = false)
-    extends Module {
-
-  val io = new Bundle {
-    val in = Vec(arbN, Decoupled(typ.cloneType)).flip
-    val out = Decoupled(typ.cloneType)
-  }
-
-  def rotateLeft[T <: Data](norm: Vec[T], rot: UInt): Vec[T] = {
-    val n = norm.size
-    Vec.tabulate(n) { i =>
-      Mux(rot < UInt(n - i), norm(UInt(i) + rot), norm(rot - UInt(n - i)))
-    }
-  }
-
-  val lockIdx = Reg(init = UInt(0, log2Up(arbN)))
-  val locked = Reg(init = Bool(false))
-
-  val choice = if (rr) {
-    PriorityMux(
-      rotateLeft(Vec(io.in.map(_.valid)), lockIdx + UInt(1)),
-      rotateLeft(Vec((0 until arbN).map(UInt(_))), lockIdx + UInt(1)))
-  } else {
-    PriorityEncoder(io.in.map(_.valid))
-  }
-
-  val chosen = Mux(locked, lockIdx, choice)
-
-  for (i <- 0 until arbN) {
-    io.in(i).ready := io.out.ready && chosen === UInt(i)
-  }
-
-  io.out.valid := io.in(chosen).valid
-  io.out.bits := io.in(chosen).bits
-}
-
-/** This locking arbiter determines when it is safe to unlock
- *  by peeking at the data */
-class JunctionsPeekingArbiter[T <: Data](
-      typ: T, arbN: Int,
-      canUnlock: T => Bool,
-      needsLock: Option[T => Bool] = None,
-      rr: Boolean = false)
-    extends JunctionsAbstractLockingArbiter(typ, arbN, rr) {
-
-  def realNeedsLock(data: T): Bool =
-    needsLock.map(_(data)).getOrElse(Bool(true))
-
-  when (io.out.fire()) {
-    when (!locked && realNeedsLock(io.out.bits)) {
-      lockIdx := choice
-      locked := Bool(true)
-    }
-    // the unlock statement takes precedent
-    when (canUnlock(io.out.bits)) {
-      locked := Bool(false)
-    }
-  }
-}
-
-/** This arbiter determines when it is safe to unlock by counting transactions */
-class JunctionsCountingArbiter[T <: Data](
-      typ: T, arbN: Int, count: Int,
-      val needsLock: Option[T => Bool] = None,
-      rr: Boolean = false)
-    extends JunctionsAbstractLockingArbiter(typ, arbN, rr) {
-
-  def realNeedsLock(data: T): Bool =
-    needsLock.map(_(data)).getOrElse(Bool(true))
-
-  // if count is 1, you should use a non-locking arbiter
-  require(count > 1, "CountingArbiter cannot have count <= 1")
-
-  val lock_ctr = Counter(count)
-
-  when (io.out.fire()) {
-    when (!locked && realNeedsLock(io.out.bits)) {
-      lockIdx := choice
-      locked := Bool(true)
-      lock_ctr.inc()
-    }
-
-    when (locked) {
-      when (lock_ctr.inc()) { locked := Bool(false) }
-    }
-  }
-}
-
-class ReorderQueueWrite[T <: Data](dType: T, tagWidth: Int) extends Bundle {
-  val data = dType.cloneType
-  val tag = UInt(width = tagWidth)
-
-  override def cloneType =
-    new ReorderQueueWrite(dType, tagWidth).asInstanceOf[this.type]
-}
-
-class ReorderEnqueueIO[T <: Data](dType: T, tagWidth: Int)
-  extends DecoupledIO(new ReorderQueueWrite(dType, tagWidth)) {
-
-  override def cloneType =
-    new ReorderEnqueueIO(dType, tagWidth).asInstanceOf[this.type]
-}
-
-class ReorderDequeueIO[T <: Data](dType: T, tagWidth: Int) extends Bundle {
-  val valid = Bool(INPUT)
-  val tag = UInt(INPUT, tagWidth)
-  val data = dType.cloneType.asOutput
-  val matches = Bool(OUTPUT)
-
-  override def cloneType =
-    new ReorderDequeueIO(dType, tagWidth).asInstanceOf[this.type]
-}
-
-class ReorderQueue[T <: Data](dType: T, tagWidth: Int, size: Option[Int] = None)
-    extends Module {
-  val io = new Bundle {
-    val enq = new ReorderEnqueueIO(dType, tagWidth).flip
-    val deq = new ReorderDequeueIO(dType, tagWidth)
-  }
-
-  val tagSpaceSize = 1 << tagWidth
-  val actualSize = size.getOrElse(tagSpaceSize)
-
-  if (tagSpaceSize > actualSize) {
-    val roq_data = Reg(Vec(actualSize, dType))
-    val roq_tags = Reg(Vec(actualSize, UInt(width = tagWidth)))
-    val roq_free = Reg(init = Vec.fill(actualSize)(Bool(true)))
-
-    val roq_enq_addr = PriorityEncoder(roq_free)
-    val roq_matches = roq_tags.zip(roq_free)
-      .map { case (tag, free) => tag === io.deq.tag && !free }
-    val roq_deq_onehot = PriorityEncoderOH(roq_matches)
-
-    io.enq.ready := roq_free.reduce(_ || _)
-    io.deq.data := Mux1H(roq_deq_onehot, roq_data)
-    io.deq.matches := roq_matches.reduce(_ || _)
-
-    when (io.enq.valid && io.enq.ready) {
-      roq_data(roq_enq_addr) := io.enq.bits.data
-      roq_tags(roq_enq_addr) := io.enq.bits.tag
-      roq_free(roq_enq_addr) := Bool(false)
-    }
-
-    when (io.deq.valid) {
-      roq_free(OHToUInt(roq_deq_onehot)) := Bool(true)
-    }
-
-    println(s"Warning - using a CAM for ReorderQueue, tagBits: ${tagWidth} size: ${actualSize}")
-  } else {
-    val roq_data = Mem(tagSpaceSize, dType)
-    val roq_free = Reg(init = Vec.fill(tagSpaceSize)(Bool(true)))
-
-    io.enq.ready := roq_free(io.enq.bits.tag)
-    io.deq.data := roq_data(io.deq.tag)
-    io.deq.matches := !roq_free(io.deq.tag)
-
-    when (io.enq.valid && io.enq.ready) {
-      roq_data(io.enq.bits.tag) := io.enq.bits.data
-      roq_free(io.enq.bits.tag) := Bool(false)
-    }
-
-    when (io.deq.valid) {
-      roq_free(io.deq.tag) := Bool(true)
-    }
-  }
-}
-
-object DecoupledHelper {
-  def apply(rvs: Bool*) = new DecoupledHelper(rvs)
-}
-
-class DecoupledHelper(val rvs: Seq[Bool]) {
-  def fire(exclude: Bool, includes: Bool*) = {
-    (rvs.filter(_ ne exclude) ++ includes).reduce(_ && _)
-  }
-}
diff --git a/src/main/scala/rocket/arbiter.scala b/src/main/scala/rocket/arbiter.scala
index 1a686d5b..de43dd6d 100644
--- a/src/main/scala/rocket/arbiter.scala
+++ b/src/main/scala/rocket/arbiter.scala
@@ -4,7 +4,7 @@ package rocket
 
 import Chisel._
 import cde.{Parameters, Field}
-import junctions.{ParameterizedBundle, DecoupledHelper}
+import util.{ParameterizedBundle, DecoupledHelper}
 
 class HellaCacheArbiter(n: Int)(implicit p: Parameters) extends Module
 {
diff --git a/src/main/scala/rocket/btb.scala b/src/main/scala/rocket/btb.scala
index 26e805d3..a4ab52de 100644
--- a/src/main/scala/rocket/btb.scala
+++ b/src/main/scala/rocket/btb.scala
@@ -3,11 +3,11 @@
 package rocket
 
 import Chisel._
-import junctions._
 import cde.{Parameters, Field}
 import Util._
 import uncore.util._
 import uncore.agents.PseudoLRU
+import util.ParameterizedBundle
 
 case object BtbKey extends Field[BtbParameters]
 
diff --git a/src/main/scala/rocket/dcache.scala b/src/main/scala/rocket/dcache.scala
index b7859881..f6b37376 100644
--- a/src/main/scala/rocket/dcache.scala
+++ b/src/main/scala/rocket/dcache.scala
@@ -127,11 +127,11 @@ class DCache(implicit p: Parameters) extends L1HellaCacheModule()(p) {
       meta.io.read <> metaReadArb.io.out
       meta.io.write <> metaWriteArb.io.out
       val s1_meta = meta.io.resp
-      val s1_hit_way = s1_meta.map(r => r.coh.isValid() && r.tag === s1_tag).asUInt
-      val s1_hit_state = ClientMetadata.onReset.fromBits(
+      val s1_meta_hit_way = s1_meta.map(r => r.coh.isValid() && r.tag === s1_tag).asUInt
+      val s1_meta_hit_state = ClientMetadata.onReset.fromBits(
         s1_meta.map(r => Mux(r.tag === s1_tag, r.coh.asUInt, UInt(0)))
         .reduce (_|_))
-      (s1_hit_way, s1_hit_state, s1_meta(s1_victim_way))
+      (s1_meta_hit_way, s1_meta_hit_state, s1_meta(s1_victim_way))
     }
   val s1_data_way = Mux(inWriteback, releaseWay, s1_hit_way)
   val s1_data = Mux1H(s1_data_way, data.io.resp) // retime into s2 if critical
diff --git a/src/main/scala/rocket/ibuf.scala b/src/main/scala/rocket/ibuf.scala
index c5a275fd..8b86793c 100644
--- a/src/main/scala/rocket/ibuf.scala
+++ b/src/main/scala/rocket/ibuf.scala
@@ -5,7 +5,7 @@ package rocket
 import Chisel._
 import Util._
 import cde.{Parameters, Field}
-import junctions._
+import util.ParameterizedBundle
 
 class Instruction(implicit val p: Parameters) extends ParameterizedBundle with HasCoreParameters {
   val pf0 = Bool() // page fault on first half of instruction
diff --git a/src/main/scala/rocket/nbdcache.scala b/src/main/scala/rocket/nbdcache.scala
index 5411faa6..12dff357 100644
--- a/src/main/scala/rocket/nbdcache.scala
+++ b/src/main/scala/rocket/nbdcache.scala
@@ -3,12 +3,12 @@
 package rocket
 
 import Chisel._
-import junctions._
 import uncore.tilelink._
 import uncore.coherence._
 import uncore.agents._
 import uncore.util._
 import uncore.constants._
+import util.{ParameterizedBundle, DecoupledHelper}
 import cde.{Parameters, Field}
 import Util._
 
@@ -44,7 +44,7 @@ trait HasL1HellaCacheParameters extends HasL1CacheParameters {
 
 abstract class L1HellaCacheModule(implicit val p: Parameters) extends Module
   with HasL1HellaCacheParameters
-abstract class L1HellaCacheBundle(implicit val p: Parameters) extends junctions.ParameterizedBundle()(p)
+abstract class L1HellaCacheBundle(implicit val p: Parameters) extends ParameterizedBundle()(p)
   with HasL1HellaCacheParameters
 
 trait HasCoreMemOp extends HasCoreParameters {
diff --git a/src/main/scala/rocket/rocket.scala b/src/main/scala/rocket/rocket.scala
index 0992b83e..6fa6893a 100644
--- a/src/main/scala/rocket/rocket.scala
+++ b/src/main/scala/rocket/rocket.scala
@@ -3,10 +3,11 @@
 package rocket
 
 import Chisel._
-import junctions._
 import uncore.devices._
 import uncore.agents.CacheName
 import uncore.constants._
+import junctions.HasAddrMapParameters
+import util.ParameterizedBundle
 import Util._
 import cde.{Parameters, Field}
 
diff --git a/src/main/scala/rocketchip/Top.scala b/src/main/scala/rocketchip/Top.scala
index 1e031ee1..7dfdb0bc 100644
--- a/src/main/scala/rocketchip/Top.scala
+++ b/src/main/scala/rocketchip/Top.scala
@@ -8,6 +8,7 @@ import junctions._
 import uncore.tilelink._
 import uncore.tilelink2.{LazyModule, LazyModuleImp}
 import uncore.devices._
+import util.ParameterizedBundle
 import rocket._
 import rocket.Util._
 import coreplex._
diff --git a/src/main/scala/uncore/agents/Agents.scala b/src/main/scala/uncore/agents/Agents.scala
index 7a74ca4c..fe10ca23 100644
--- a/src/main/scala/uncore/agents/Agents.scala
+++ b/src/main/scala/uncore/agents/Agents.scala
@@ -4,7 +4,8 @@ package uncore.agents
 
 import Chisel._
 import cde.{Parameters, Field}
-import junctions._
+import junctions.PAddrBits
+import util.ParameterizedBundle
 import uncore.tilelink._
 import uncore.converters._
 import uncore.coherence._
@@ -41,7 +42,7 @@ trait HasCoherenceAgentParameters {
 
 abstract class CoherenceAgentModule(implicit val p: Parameters) extends Module
   with HasCoherenceAgentParameters
-abstract class CoherenceAgentBundle(implicit val p: Parameters) extends junctions.ParameterizedBundle()(p)
+abstract class CoherenceAgentBundle(implicit val p: Parameters) extends ParameterizedBundle()(p)
    with HasCoherenceAgentParameters
 
 trait HasCoherenceAgentWiringHelpers {
diff --git a/src/main/scala/uncore/agents/Cache.scala b/src/main/scala/uncore/agents/Cache.scala
index 1c9e29e2..77603007 100644
--- a/src/main/scala/uncore/agents/Cache.scala
+++ b/src/main/scala/uncore/agents/Cache.scala
@@ -4,7 +4,8 @@ package uncore.agents
 
 import Chisel._
 import scala.reflect.ClassTag
-import junctions._
+import junctions.PAddrBits
+import util.ParameterizedBundle
 import uncore.util.AMOALU
 import uncore.coherence._
 import uncore.tilelink._
diff --git a/src/main/scala/uncore/agents/Trackers.scala b/src/main/scala/uncore/agents/Trackers.scala
index 96fe69e2..d7879530 100644
--- a/src/main/scala/uncore/agents/Trackers.scala
+++ b/src/main/scala/uncore/agents/Trackers.scala
@@ -7,7 +7,7 @@ import uncore.coherence._
 import uncore.tilelink._
 import uncore.util._
 import uncore.util._
-import junctions._
+import util.ParameterizedBundle
 import cde.{Field, Parameters}
 import scala.math.max
 
diff --git a/src/main/scala/uncore/converters/Nasti.scala b/src/main/scala/uncore/converters/Nasti.scala
index ecf75467..0743b488 100644
--- a/src/main/scala/uncore/converters/Nasti.scala
+++ b/src/main/scala/uncore/converters/Nasti.scala
@@ -2,6 +2,7 @@ package uncore.converters
 
 import Chisel._
 import junctions._
+import util.{ReorderQueue, DecoupledHelper}
 import uncore.tilelink._
 import uncore.constants._
 import cde.Parameters
diff --git a/src/main/scala/uncore/converters/Tilelink.scala b/src/main/scala/uncore/converters/Tilelink.scala
index d77bfa65..f14498b7 100644
--- a/src/main/scala/uncore/converters/Tilelink.scala
+++ b/src/main/scala/uncore/converters/Tilelink.scala
@@ -1,7 +1,8 @@
 package uncore.converters
 
 import Chisel._
-import junctions._
+import util.{ReorderQueue, DecoupledHelper}
+import junctions.PAddrBits
 import uncore.tilelink._
 import uncore.util._
 import uncore.constants._
diff --git a/src/main/scala/uncore/devices/Debug.scala b/src/main/scala/uncore/devices/Debug.scala
index 6714a741..28e7d9bf 100644
--- a/src/main/scala/uncore/devices/Debug.scala
+++ b/src/main/scala/uncore/devices/Debug.scala
@@ -3,9 +3,10 @@
 package uncore.devices
 
 import Chisel._
+import junctions._
 import uncore.tilelink._
 import uncore.util._
-import junctions._
+import util.ParameterizedBundle
 import cde.{Parameters, Config, Field}
 
 // *****************************************
diff --git a/src/main/scala/uncore/tilelink/Definitions.scala b/src/main/scala/uncore/tilelink/Definitions.scala
index c52fb0a2..e32b9bdd 100644
--- a/src/main/scala/uncore/tilelink/Definitions.scala
+++ b/src/main/scala/uncore/tilelink/Definitions.scala
@@ -90,7 +90,7 @@ trait HasTileLinkParameters {
 
 abstract class TLModule(implicit val p: Parameters) extends Module
   with HasTileLinkParameters
-abstract class TLBundle(implicit val p: Parameters) extends junctions.ParameterizedBundle()(p)
+abstract class TLBundle(implicit val p: Parameters) extends util.ParameterizedBundle()(p)
   with HasTileLinkParameters
 
 /** Base trait for all TileLink channels */
diff --git a/src/main/scala/uncore/tilelink/Network.scala b/src/main/scala/uncore/tilelink/Network.scala
index 1c094013..1a572e78 100644
--- a/src/main/scala/uncore/tilelink/Network.scala
+++ b/src/main/scala/uncore/tilelink/Network.scala
@@ -61,7 +61,7 @@ class BasicCrossbar[T <: Data](conf: CrossbarConfig[T]) extends AbstractCrossbar
 
 abstract class LogicalNetwork extends Module
 
-class LogicalHeader(implicit p: Parameters) extends junctions.ParameterizedBundle()(p) {
+class LogicalHeader(implicit p: Parameters) extends util.ParameterizedBundle()(p) {
   val src = UInt(width = p(LNHeaderBits))
   val dst = UInt(width = p(LNHeaderBits))
 }
diff --git a/src/main/scala/util/Arbiters.scala b/src/main/scala/util/Arbiters.scala
new file mode 100644
index 00000000..b5518720
--- /dev/null
+++ b/src/main/scala/util/Arbiters.scala
@@ -0,0 +1,93 @@
+package util
+import Chisel._
+import cde.Parameters
+
+/** A generalized locking RR arbiter that addresses the limitations of the
+ *  version in the Chisel standard library */
+abstract class HellaLockingArbiter[T <: Data](typ: T, arbN: Int, rr: Boolean = false)
+    extends Module {
+
+  val io = new Bundle {
+    val in = Vec(arbN, Decoupled(typ.cloneType)).flip
+    val out = Decoupled(typ.cloneType)
+  }
+
+  def rotateLeft[T <: Data](norm: Vec[T], rot: UInt): Vec[T] = {
+    val n = norm.size
+    Vec.tabulate(n) { i =>
+      Mux(rot < UInt(n - i), norm(UInt(i) + rot), norm(rot - UInt(n - i)))
+    }
+  }
+
+  val lockIdx = Reg(init = UInt(0, log2Up(arbN)))
+  val locked = Reg(init = Bool(false))
+
+  val choice = if (rr) {
+    PriorityMux(
+      rotateLeft(Vec(io.in.map(_.valid)), lockIdx + UInt(1)),
+      rotateLeft(Vec((0 until arbN).map(UInt(_))), lockIdx + UInt(1)))
+  } else {
+    PriorityEncoder(io.in.map(_.valid))
+  }
+
+  val chosen = Mux(locked, lockIdx, choice)
+
+  for (i <- 0 until arbN) {
+    io.in(i).ready := io.out.ready && chosen === UInt(i)
+  }
+
+  io.out.valid := io.in(chosen).valid
+  io.out.bits := io.in(chosen).bits
+}
+
+/** This locking arbiter determines when it is safe to unlock
+ *  by peeking at the data */
+class HellaPeekingArbiter[T <: Data](
+      typ: T, arbN: Int,
+      canUnlock: T => Bool,
+      needsLock: Option[T => Bool] = None,
+      rr: Boolean = false)
+    extends HellaLockingArbiter(typ, arbN, rr) {
+
+  def realNeedsLock(data: T): Bool =
+    needsLock.map(_(data)).getOrElse(Bool(true))
+
+  when (io.out.fire()) {
+    when (!locked && realNeedsLock(io.out.bits)) {
+      lockIdx := choice
+      locked := Bool(true)
+    }
+    // the unlock statement takes precedent
+    when (canUnlock(io.out.bits)) {
+      locked := Bool(false)
+    }
+  }
+}
+
+/** This arbiter determines when it is safe to unlock by counting transactions */
+class HellaCountingArbiter[T <: Data](
+      typ: T, arbN: Int, count: Int,
+      val needsLock: Option[T => Bool] = None,
+      rr: Boolean = false)
+    extends HellaLockingArbiter(typ, arbN, rr) {
+
+  def realNeedsLock(data: T): Bool =
+    needsLock.map(_(data)).getOrElse(Bool(true))
+
+  // if count is 1, you should use a non-locking arbiter
+  require(count > 1, "CountingArbiter cannot have count <= 1")
+
+  val lock_ctr = Counter(count)
+
+  when (io.out.fire()) {
+    when (!locked && realNeedsLock(io.out.bits)) {
+      lockIdx := choice
+      locked := Bool(true)
+      lock_ctr.inc()
+    }
+
+    when (locked) {
+      when (lock_ctr.inc()) { locked := Bool(false) }
+    }
+  }
+}
diff --git a/src/main/scala/util/HellaQueue.scala b/src/main/scala/util/HellaQueue.scala
new file mode 100644
index 00000000..acaa800b
--- /dev/null
+++ b/src/main/scala/util/HellaQueue.scala
@@ -0,0 +1,55 @@
+package util
+
+import Chisel._
+import cde.Parameters
+
+class HellaFlowQueue[T <: Data](val entries: Int)(data: => T) extends Module {
+  val io = new QueueIO(data, entries)
+  require(entries > 1)
+
+  val do_flow = Wire(Bool())
+  val do_enq = io.enq.fire() && !do_flow
+  val do_deq = io.deq.fire() && !do_flow
+
+  val maybe_full = Reg(init=Bool(false))
+  val enq_ptr = Counter(do_enq, entries)._1
+  val (deq_ptr, deq_done) = Counter(do_deq, entries)
+  when (do_enq =/= do_deq) { maybe_full := do_enq }
+
+  val ptr_match = enq_ptr === deq_ptr
+  val empty = ptr_match && !maybe_full
+  val full = ptr_match && maybe_full
+  val atLeastTwo = full || enq_ptr - deq_ptr >= UInt(2)
+  do_flow := empty && io.deq.ready
+
+  val ram = SeqMem(entries, data)
+  when (do_enq) { ram.write(enq_ptr, io.enq.bits) }
+
+  val ren = io.deq.ready && (atLeastTwo || !io.deq.valid && !empty)
+  val raddr = Mux(io.deq.valid, Mux(deq_done, UInt(0), deq_ptr + UInt(1)), deq_ptr)
+  val ram_out_valid = Reg(next = ren)
+
+  io.deq.valid := Mux(empty, io.enq.valid, ram_out_valid)
+  io.enq.ready := !full
+  io.deq.bits := Mux(empty, io.enq.bits, ram.read(raddr, ren))
+}
+
+class HellaQueue[T <: Data](val entries: Int)(data: => T) extends Module {
+  val io = new QueueIO(data, entries)
+
+  val fq = Module(new HellaFlowQueue(entries)(data))
+  fq.io.enq <> io.enq
+  io.deq <> Queue(fq.io.deq, 1, pipe = true)
+}
+
+object HellaQueue {
+  def apply[T <: Data](enq: DecoupledIO[T], entries: Int) = {
+    val q = Module((new HellaQueue(entries)) { enq.bits })
+    q.io.enq.valid := enq.valid // not using <> so that override is allowed
+    q.io.enq.bits := enq.bits
+    enq.ready := q.io.enq.ready
+    q.io.deq
+  }
+}
+
+
diff --git a/src/main/scala/util/Misc.scala b/src/main/scala/util/Misc.scala
new file mode 100644
index 00000000..80fd8107
--- /dev/null
+++ b/src/main/scala/util/Misc.scala
@@ -0,0 +1,28 @@
+package util
+
+import Chisel._
+import cde.Parameters
+
+class ParameterizedBundle(implicit p: Parameters) extends Bundle {
+  override def cloneType = {
+    try {
+      this.getClass.getConstructors.head.newInstance(p).asInstanceOf[this.type]
+    } catch {
+      case e: java.lang.IllegalArgumentException =>
+        throwException("Unable to use ParamaterizedBundle.cloneType on " +
+                       this.getClass + ", probably because " + this.getClass +
+                       "() takes more than one argument.  Consider overriding " +
+                       "cloneType() on " + this.getClass, e)
+    }
+  }
+}
+
+object DecoupledHelper {
+  def apply(rvs: Bool*) = new DecoupledHelper(rvs)
+}
+
+class DecoupledHelper(val rvs: Seq[Bool]) {
+  def fire(exclude: Bool, includes: Bool*) = {
+    (rvs.filter(_ ne exclude) ++ includes).reduce(_ && _)
+  }
+}
diff --git a/src/main/scala/util/ReorderQueue.scala b/src/main/scala/util/ReorderQueue.scala
new file mode 100644
index 00000000..458001af
--- /dev/null
+++ b/src/main/scala/util/ReorderQueue.scala
@@ -0,0 +1,85 @@
+package util
+
+import Chisel._
+import cde.Parameters
+
+class ReorderQueueWrite[T <: Data](dType: T, tagWidth: Int) extends Bundle {
+  val data = dType.cloneType
+  val tag = UInt(width = tagWidth)
+
+  override def cloneType =
+    new ReorderQueueWrite(dType, tagWidth).asInstanceOf[this.type]
+}
+
+class ReorderEnqueueIO[T <: Data](dType: T, tagWidth: Int)
+  extends DecoupledIO(new ReorderQueueWrite(dType, tagWidth)) {
+
+  override def cloneType =
+    new ReorderEnqueueIO(dType, tagWidth).asInstanceOf[this.type]
+}
+
+class ReorderDequeueIO[T <: Data](dType: T, tagWidth: Int) extends Bundle {
+  val valid = Bool(INPUT)
+  val tag = UInt(INPUT, tagWidth)
+  val data = dType.cloneType.asOutput
+  val matches = Bool(OUTPUT)
+
+  override def cloneType =
+    new ReorderDequeueIO(dType, tagWidth).asInstanceOf[this.type]
+}
+
+class ReorderQueue[T <: Data](dType: T, tagWidth: Int, size: Option[Int] = None)
+    extends Module {
+  val io = new Bundle {
+    val enq = new ReorderEnqueueIO(dType, tagWidth).flip
+    val deq = new ReorderDequeueIO(dType, tagWidth)
+  }
+
+  val tagSpaceSize = 1 << tagWidth
+  val actualSize = size.getOrElse(tagSpaceSize)
+
+  if (tagSpaceSize > actualSize) {
+    val roq_data = Reg(Vec(actualSize, dType))
+    val roq_tags = Reg(Vec(actualSize, UInt(width = tagWidth)))
+    val roq_free = Reg(init = Vec.fill(actualSize)(Bool(true)))
+
+    val roq_enq_addr = PriorityEncoder(roq_free)
+    val roq_matches = roq_tags.zip(roq_free)
+      .map { case (tag, free) => tag === io.deq.tag && !free }
+    val roq_deq_onehot = PriorityEncoderOH(roq_matches)
+
+    io.enq.ready := roq_free.reduce(_ || _)
+    io.deq.data := Mux1H(roq_deq_onehot, roq_data)
+    io.deq.matches := roq_matches.reduce(_ || _)
+
+    when (io.enq.valid && io.enq.ready) {
+      roq_data(roq_enq_addr) := io.enq.bits.data
+      roq_tags(roq_enq_addr) := io.enq.bits.tag
+      roq_free(roq_enq_addr) := Bool(false)
+    }
+
+    when (io.deq.valid) {
+      roq_free(OHToUInt(roq_deq_onehot)) := Bool(true)
+    }
+
+    println(s"Warning - using a CAM for ReorderQueue, tagBits: ${tagWidth} size: ${actualSize}")
+  } else {
+    val roq_data = Mem(tagSpaceSize, dType)
+    val roq_free = Reg(init = Vec.fill(tagSpaceSize)(Bool(true)))
+
+    io.enq.ready := roq_free(io.enq.bits.tag)
+    io.deq.data := roq_data(io.deq.tag)
+    io.deq.matches := !roq_free(io.deq.tag)
+
+    when (io.enq.valid && io.enq.ready) {
+      roq_data(io.enq.bits.tag) := io.enq.bits.data
+      roq_free(io.enq.bits.tag) := Bool(false)
+    }
+
+    when (io.deq.valid) {
+      roq_free(io.deq.tag) := Bool(true)
+    }
+  }
+}
+
+