From ecdfb528c5d8491ff5a645b33d1204d880ddcdef Mon Sep 17 00:00:00 2001
From: "Wesley W. Terpstra" <wesley@sifive.com>
Date: Wed, 7 Sep 2016 13:55:22 -0700
Subject: [PATCH 01/10] crossing: refactor AsyncDecoupled to provide
 AsyncDecoupledCrossing with no clock domain

---
 src/main/scala/junctions/crossing.scala       | 86 ++++++++-----------
 src/main/scala/junctions/nasti.scala          | 45 +++++-----
 .../scala/rocketchip/DebugTransport.scala     | 27 +++---
 src/main/scala/uncore/devices/Debug.scala     | 30 ++++---
 4 files changed, 91 insertions(+), 97 deletions(-)

diff --git a/src/main/scala/junctions/crossing.scala b/src/main/scala/junctions/crossing.scala
index 0f71362e..802504c0 100644
--- a/src/main/scala/junctions/crossing.scala
+++ b/src/main/scala/junctions/crossing.scala
@@ -1,13 +1,13 @@
 package junctions
 import Chisel._
 
-class Crossing[T <: Data](gen: T, enq_sync: Boolean, deq_sync: Boolean) extends Bundle {
+class Crossing[T <: Data](gen: T) extends Bundle {
     val enq = Decoupled(gen).flip()
     val deq = Decoupled(gen)
-    val enq_clock = if (enq_sync) Some(Clock(INPUT)) else None
-    val deq_clock = if (deq_sync) Some(Clock(INPUT)) else None
-    val enq_reset = if (enq_sync) Some(Bool(INPUT))  else None
-    val deq_reset = if (deq_sync) Some(Bool(INPUT))  else None
+    val enq_clock = Clock(INPUT)
+    val deq_clock = Clock(INPUT)
+    val enq_reset = Bool(INPUT)
+    val deq_reset = Bool(INPUT)
 }
 
 // Output is 1 for one cycle after any edge of 'in'
@@ -87,11 +87,11 @@ class AsyncHandshakeSink[T <: Data](gen: T, sync: Int, clock: Clock, reset: Bool
 }
 
 class AsyncHandshake[T <: Data](gen: T, sync: Int = 2) extends Module {
-  val io = new Crossing(gen, true, true)
+  val io = new Crossing(gen)
   require (sync >= 2)
 
-  val source = Module(new AsyncHandshakeSource(gen, sync, io.enq_clock.get, io.enq_reset.get))
-  val sink   = Module(new AsyncHandshakeSink  (gen, sync, io.deq_clock.get, io.deq_reset.get))
+  val source = Module(new AsyncHandshakeSource(gen, sync, io.enq_clock, io.enq_reset))
+  val sink   = Module(new AsyncHandshakeSink  (gen, sync, io.deq_clock, io.deq_reset))
 
   source.io.enq <> io.enq
   io.deq <> sink.io.deq
@@ -101,50 +101,38 @@ class AsyncHandshake[T <: Data](gen: T, sync: Int = 2) extends Module {
   source.io.pop := sink.io.pop
 }
 
-class AsyncDecoupledTo[T <: Data](gen: T, depth: Int = 0, sync: Int = 2) extends Module {
-  val io = new Crossing(gen, false, true)
+class AsyncScope extends Module { val io = new Bundle }
+object AsyncScope { def apply() = Module(new AsyncScope) }
 
-  // !!! if depth == 0 { use Handshake } else { use AsyncFIFO }
-  val crossing = Module(new AsyncHandshake(gen, sync)).io
-  crossing.enq_clock.get := clock
-  crossing.enq_reset.get := reset
-  crossing.enq <> io.enq
-  crossing.deq_clock.get := io.deq_clock.get
-  crossing.deq_reset.get := io.deq_reset.get
-  io.deq <> crossing.deq
-}
-
-object AsyncDecoupledTo {
-  // source is in our clock domain, output is in the 'to' clock domain
-  def apply[T <: Data](to_clock: Clock, to_reset: Bool, source: DecoupledIO[T], depth: Int = 0, sync: Int = 2): DecoupledIO[T] = {
-    val to = Module(new AsyncDecoupledTo(source.bits, depth, sync))
-    to.io.deq_clock.get := to_clock
-    to.io.deq_reset.get := to_reset
-    to.io.enq <> source
-    to.io.deq
+object AsyncDecoupledCrossing
+{
+  // takes from_source from the 'from' clock domain and puts it into the 'to' clock domain
+  def apply[T <: Data](from_clock: Clock, from_reset: Bool, from_source: DecoupledIO[T], to_clock: Clock, to_reset: Bool, depth: Int = 3, sync: Int = 2): DecoupledIO[T] = {
+    // !!! if depth == 0 { use Handshake } else { use AsyncFIFO }
+    val crossing = Module(new AsyncHandshake(from_source.bits, sync)).io
+    crossing.enq_clock := from_clock
+    crossing.enq_reset := from_reset
+    crossing.enq       <> from_source
+    crossing.deq_clock := to_clock
+    crossing.deq_reset := to_reset
+    crossing.deq
   }
 }
 
-class AsyncDecoupledFrom[T <: Data](gen: T, depth: Int = 0, sync: Int = 2) extends Module {
-  val io = new Crossing(gen, true, false)
-
-  // !!! if depth == 0 { use Handshake } else { use AsyncFIFO }
-  val crossing = Module(new AsyncHandshake(gen, sync)).io
-  crossing.enq_clock.get := io.enq_clock.get
-  crossing.enq_reset.get := io.enq_reset.get
-  crossing.enq <> io.enq
-  crossing.deq_clock.get := clock
-  crossing.deq_reset.get := reset
-  io.deq <> crossing.deq
-}
-
-object AsyncDecoupledFrom {
-  // source is in the 'from' clock domain, output is in our clock domain
-  def apply[T <: Data](from_clock: Clock, from_reset: Bool, source: DecoupledIO[T], depth: Int = 0, sync: Int = 2): DecoupledIO[T] = {
-    val from = Module(new AsyncDecoupledFrom(source.bits, depth, sync))
-    from.io.enq_clock.get := from_clock
-    from.io.enq_reset.get := from_reset
-    from.io.enq <> source
-    from.io.deq
+object AsyncDecoupledTo
+{
+  // takes source from your clock domain and puts it into the 'to' clock domain
+  def apply[T <: Data](to_clock: Clock, to_reset: Bool, source: DecoupledIO[T], depth: Int = 3, sync: Int = 2): DecoupledIO[T] = {
+    val scope = AsyncScope()
+    AsyncDecoupledCrossing(scope.clock, scope.reset, source, to_clock, to_reset, depth, sync)
+  }
+}
+
+object AsyncDecoupledFrom
+{
+  // takes from_source from the 'from' clock domain and puts it into your clock domain
+  def apply[T <: Data](from_clock: Clock, from_reset: Bool, from_source: DecoupledIO[T], depth: Int = 3, sync: Int = 2): DecoupledIO[T] = {
+    val scope = AsyncScope()
+    AsyncDecoupledCrossing(from_clock, from_reset, from_source, scope.clock, scope.reset, depth, sync)
   }
 }
diff --git a/src/main/scala/junctions/nasti.scala b/src/main/scala/junctions/nasti.scala
index 41251ac3..3ae1fbc2 100644
--- a/src/main/scala/junctions/nasti.scala
+++ b/src/main/scala/junctions/nasti.scala
@@ -706,32 +706,33 @@ class NastiMemoryDemux(nRoutes: Int)(implicit p: Parameters) extends NastiModule
   }
 }
 
+object AsyncNastiCrossing {
+  // takes from_source from the 'from' clock domain to the 'to' clock domain
+  def apply(from_clock: Clock, from_reset: Bool, from_source: NastiIO, to_clock: Clock, to_reset: Bool, depth: Int = 3, sync: Int = 2) = {
+    val to_sink = Wire(new NastiIO()(from_source.p))
+
+    to_sink.aw <> AsyncDecoupledCrossing(from_clock, from_reset, from_source.aw, to_clock, to_reset, depth, sync)
+    to_sink.ar <> AsyncDecoupledCrossing(from_clock, from_reset, from_source.ar, to_clock, to_reset, depth, sync)
+    to_sink.w  <> AsyncDecoupledCrossing(from_clock, from_reset, from_source.w,  to_clock, to_reset, depth, sync)
+    from_source.b <> AsyncDecoupledCrossing(to_clock, to_reset, to_sink.b, from_clock, from_reset, depth, sync)
+    from_source.r <> AsyncDecoupledCrossing(to_clock, to_reset, to_sink.r, from_clock, from_reset, depth, sync)
+
+    to_sink // is now to_source
+  }
+}
+
 object AsyncNastiTo {
-  // source(master) is in our clock domain, output is in the 'to' clock domain
-  def apply[T <: Data](to_clock: Clock, to_reset: Bool, source: NastiIO, depth: Int = 3, sync: Int = 2)(implicit p: Parameters): NastiIO = {
-    val sink = Wire(new NastiIO)
-
-    sink.aw <> AsyncDecoupledTo(to_clock, to_reset, source.aw, depth, sync)
-    sink.ar <> AsyncDecoupledTo(to_clock, to_reset, source.ar, depth, sync)
-    sink.w  <> AsyncDecoupledTo(to_clock, to_reset, source.w,  depth, sync)
-    source.b <> AsyncDecoupledFrom(to_clock, to_reset, sink.b, depth, sync)
-    source.r <> AsyncDecoupledFrom(to_clock, to_reset, sink.r, depth, sync)
-
-    sink
+  // takes source from your clock domain and puts it into the 'to' clock domain
+  def apply(to_clock: Clock, to_reset: Bool, source: NastiIO, depth: Int = 3, sync: Int = 2): NastiIO = {
+    val scope = AsyncScope()
+    AsyncNastiCrossing(scope.clock, scope.reset, source, to_clock, to_reset, depth, sync)
   }
 }
 
 object AsyncNastiFrom {
-  // source(master) is in the 'from' clock domain, output is in our clock domain
-  def apply[T <: Data](from_clock: Clock, from_reset: Bool, source: NastiIO, depth: Int = 3, sync: Int = 2)(implicit p: Parameters): NastiIO = {
-    val sink = Wire(new NastiIO)
-
-    sink.aw <> AsyncDecoupledFrom(from_clock, from_reset, source.aw, depth, sync)
-    sink.ar <> AsyncDecoupledFrom(from_clock, from_reset, source.ar, depth, sync)
-    sink.w  <> AsyncDecoupledFrom(from_clock, from_reset, source.w,  depth, sync)
-    source.b <> AsyncDecoupledTo(from_clock, from_reset, sink.b, depth, sync)
-    source.r <> AsyncDecoupledTo(from_clock, from_reset, sink.r, depth, sync)
-
-    sink
+  // takes from_source from the 'from' clock domain and puts it into your clock domain
+  def apply(from_clock: Clock, from_reset: Bool, from_source: NastiIO, depth: Int = 3, sync: Int = 2): NastiIO = {
+    val scope = AsyncScope()
+    AsyncNastiCrossing(from_clock, from_reset, from_source, scope.clock, scope.reset, depth, sync)
   }
 }
diff --git a/src/main/scala/rocketchip/DebugTransport.scala b/src/main/scala/rocketchip/DebugTransport.scala
index 92fd0f29..846a6500 100644
--- a/src/main/scala/rocketchip/DebugTransport.scala
+++ b/src/main/scala/rocketchip/DebugTransport.scala
@@ -86,19 +86,19 @@ class JtagDTMWithSync(implicit val p: Parameters) extends Module {
   } else {
     val req_sync  = Module (new AsyncMailbox())
     val resp_sync = Module (new AsyncMailbox())
-    req_sync.io.enq             := jtag_dtm.io.dtm_req
-    req_sync.io.enq_clock.get   := io.jtag.TCK
-    req_sync.io.enq_reset.get   := io.jtag.TRST
-    req_sync.io.deq_clock.get   := clock
-    req_sync.io.deq_reset.get   := reset
-    dtm_req                     := req_sync.io.deq
+    req_sync.io.enq         := jtag_dtm.io.dtm_req
+    req_sync.io.enq_clock   := io.jtag.TCK
+    req_sync.io.enq_reset   := io.jtag.TRST
+    req_sync.io.deq_clock   := clock
+    req_sync.io.deq_reset   := reset
+    dtm_req                 := req_sync.io.deq
 
-    jtag_dtm.io.dtm_resp        := resp_sync.io.deq
-    resp_sync.io.deq_clock.get  := io.jtag.TCK
-    resp_sync.io.deq_reset.get  := io.jtag.TRST
-    resp_sync.io.enq_clock.get  := clock
-    resp_sync.io.enq_reset.get  := reset
-    resp_sync.io.enq            := dtm_resp
+    jtag_dtm.io.dtm_resp    := resp_sync.io.deq
+    resp_sync.io.deq_clock  := io.jtag.TCK
+    resp_sync.io.deq_reset  := io.jtag.TRST
+    resp_sync.io.enq_clock  := clock
+    resp_sync.io.enq_reset  := reset
+    resp_sync.io.enq        := dtm_resp
   }
 }
 
@@ -121,6 +121,5 @@ class AsyncMailbox extends BlackBox {
   // this mailbox just has a fixed width of 64 bits, which is enough
   // for our specific purpose here.
 
-  val io = new Crossing(UInt(width=64), true, true)
-
+  val io = new Crossing(UInt(width=64))
 }
diff --git a/src/main/scala/uncore/devices/Debug.scala b/src/main/scala/uncore/devices/Debug.scala
index 7443c9ac..310bdc68 100644
--- a/src/main/scala/uncore/devices/Debug.scala
+++ b/src/main/scala/uncore/devices/Debug.scala
@@ -982,22 +982,28 @@ class DebugModule ()(implicit val p:cde.Parameters)
 
 }
 
+object AsyncDebugBusCrossing {
+  // takes from_source from the 'from' clock domain to the 'to' clock domain
+  def apply(from_clock: Clock, from_reset: Bool, from_source: DebugBusIO, to_clock: Clock, to_reset: Bool, depth: Int = 3, sync: Int = 2) = {
+    val to_sink = Wire(new DebugBusIO()(from_source.p))
+    to_sink.req <> AsyncDecoupledCrossing(from_clock, from_reset, from_source.req, to_clock, to_reset, depth, sync)
+    from_source.resp <> AsyncDecoupledCrossing(to_clock, to_reset, to_sink.resp, from_clock, from_reset, depth, sync)
+    to_sink // is now to_source
+  }
+}
+
 object AsyncDebugBusFrom { // OutsideClockDomain
-  def apply(from_clock: Clock, from_reset: Bool, source: DebugBusIO, depth: Int = 0, sync: Int = 2)(implicit p: Parameters): DebugBusIO = {
-    val sink = Wire(new DebugBusIO)
-    sink.req <> AsyncDecoupledFrom(from_clock, from_reset, source.req)
-    source.resp <> AsyncDecoupledTo(from_clock, from_reset, sink.resp)
-    sink
+  // takes from_source from the 'from' clock domain and puts it into your clock domain
+  def apply(from_clock: Clock, from_reset: Bool, from_source: DebugBusIO, depth: Int = 0, sync: Int = 2): DebugBusIO = {
+    val scope = AsyncScope()
+    AsyncDebugBusCrossing(from_clock, from_reset, from_source, scope.clock, scope.reset, depth, sync)
   }
 }
 
 object AsyncDebugBusTo { // OutsideClockDomain
-  def apply(to_clock: Clock, to_reset: Bool, source: DebugBusIO, depth: Int = 0, sync: Int = 2)(implicit p: Parameters): DebugBusIO = {
-    val sink = Wire(new DebugBusIO)
-    sink.req <> AsyncDecoupledTo(to_clock, to_reset, source.req)
-    source.resp <> AsyncDecoupledFrom(to_clock, to_reset, sink.resp)
-    sink
+  // takes source from your clock domain and puts it into the 'to' clock domain
+  def apply(to_clock: Clock, to_reset: Bool, source: DebugBusIO, depth: Int = 0, sync: Int = 2): DebugBusIO = {
+    val scope = AsyncScope()
+    AsyncDebugBusCrossing(scope.clock, scope.reset, source, to_clock, to_reset, depth, sync)
   }
 }
-
-

From 8142406d2e5a0138c8b520ecc4048e7a200c3ec2 Mon Sep 17 00:00:00 2001
From: "Wesley W. Terpstra" <wesley@sifive.com>
Date: Tue, 13 Sep 2016 15:34:56 -0700
Subject: [PATCH 02/10] junctions: refactor the Crossing type

---
 src/main/scala/junctions/crossing.scala       | 24 ++++++++++++-------
 .../scala/rocketchip/DebugTransport.scala     |  2 +-
 2 files changed, 16 insertions(+), 10 deletions(-)

diff --git a/src/main/scala/junctions/crossing.scala b/src/main/scala/junctions/crossing.scala
index 802504c0..f0db6f86 100644
--- a/src/main/scala/junctions/crossing.scala
+++ b/src/main/scala/junctions/crossing.scala
@@ -1,13 +1,19 @@
 package junctions
 import Chisel._
 
-class Crossing[T <: Data](gen: T) extends Bundle {
-    val enq = Decoupled(gen).flip()
-    val deq = Decoupled(gen)
-    val enq_clock = Clock(INPUT)
-    val deq_clock = Clock(INPUT)
-    val enq_reset = Bool(INPUT)
-    val deq_reset = Bool(INPUT)
+class CrossingIO[T <: Data](gen: T) extends Bundle {
+  // Enqueue clock domain
+  val enq_clock = Clock(INPUT)
+  val enq_reset = Bool(INPUT) // synchronously deasserted wrt. enq_clock
+  val enq = Decoupled(gen).flip()
+  // Dequeue clock domain
+  val deq_clock = Clock(INPUT)
+  val deq_reset = Bool(INPUT) // synchronously deasserted wrt. deq_clock
+  val deq = Decoupled(gen)
+}
+
+abstract class Crossing[T <: Data] extends Module {
+  val io: CrossingIO[T]
 }
 
 // Output is 1 for one cycle after any edge of 'in'
@@ -86,8 +92,8 @@ class AsyncHandshakeSink[T <: Data](gen: T, sync: Int, clock: Clock, reset: Bool
   }
 }
 
-class AsyncHandshake[T <: Data](gen: T, sync: Int = 2) extends Module {
-  val io = new Crossing(gen)
+class AsyncHandshake[T <: Data](gen: T, sync: Int = 2) extends Crossing[T] {
+  val io = new CrossingIO(gen)
   require (sync >= 2)
 
   val source = Module(new AsyncHandshakeSource(gen, sync, io.enq_clock, io.enq_reset))
diff --git a/src/main/scala/rocketchip/DebugTransport.scala b/src/main/scala/rocketchip/DebugTransport.scala
index 846a6500..1371a16d 100644
--- a/src/main/scala/rocketchip/DebugTransport.scala
+++ b/src/main/scala/rocketchip/DebugTransport.scala
@@ -121,5 +121,5 @@ class AsyncMailbox extends BlackBox {
   // this mailbox just has a fixed width of 64 bits, which is enough
   // for our specific purpose here.
 
-  val io = new Crossing(UInt(width=64))
+  val io = new CrossingIO(UInt(width=64))
 }

From d75f9d6a34d015674bfcff53eafff9d6d538c6b9 Mon Sep 17 00:00:00 2001
From: "Wesley W. Terpstra" <wesley@sifive.com>
Date: Tue, 13 Sep 2016 15:30:09 -0700
Subject: [PATCH 03/10] junctions: add an AsyncQueue

---
 src/main/scala/junctions/asyncqueue.scala | 88 +++++++++++++++++++++++
 1 file changed, 88 insertions(+)
 create mode 100644 src/main/scala/junctions/asyncqueue.scala

diff --git a/src/main/scala/junctions/asyncqueue.scala b/src/main/scala/junctions/asyncqueue.scala
new file mode 100644
index 00000000..585fa9e9
--- /dev/null
+++ b/src/main/scala/junctions/asyncqueue.scala
@@ -0,0 +1,88 @@
+// See LICENSE for license details.
+
+package junctions
+import Chisel._
+
+object GrayCounter {
+  def apply(bits: Int, increment: Bool = Bool(true)): UInt = {
+    val binary = RegInit(UInt(0, width = bits))
+    val incremented = binary + increment.asUInt()
+    binary := incremented
+    incremented ^ (incremented >> UInt(1))
+  }
+}
+
+object AsyncGrayCounter {
+  def apply(in: UInt, sync: Int): UInt = {
+    val syncv = RegInit(Vec.fill(sync){UInt(0, width = in.getWidth)})
+    syncv.last := in
+    (syncv.init zip syncv.tail).foreach { case (sink, source) => sink := source }
+    syncv(0)
+  }
+}
+
+class AsyncQueueSource[T <: Data](gen: T, depth: Int, sync: Int, clockIn: Clock, resetIn: Bool)
+    extends Module(_clock = clockIn, _reset = resetIn) {
+  val bits = log2Ceil(depth)
+  val io = new Bundle {
+    // These come from the source domain
+    val enq  = Decoupled(gen).flip()
+    // These cross to the sink clock domain
+    val ridx = UInt(INPUT,  width = bits+1)
+    val widx = UInt(OUTPUT, width = bits+1)
+    val mem  = Vec(depth, gen).asOutput
+  }
+
+  val mem = Reg(Vec(depth, gen))
+  val widx = GrayCounter(bits+1, io.enq.fire())
+  val ridx = AsyncGrayCounter(io.ridx, sync)
+  val ready = widx =/= (ridx ^ UInt(depth | depth >> 1))
+
+  val index = if (depth == 1) UInt(0) else io.widx(bits-1, 0) ^ (io.widx(bits, bits) << (bits-1))
+  when (io.enq.fire() && !reset) { mem(index) := io.enq.bits }
+  io.enq.ready := RegNext(ready, Bool(false))
+  io.widx := RegNext(widx, UInt(0))
+  io.mem := mem
+}
+
+class AsyncQueueSink[T <: Data](gen: T, depth: Int, sync: Int, clockIn: Clock, resetIn: Bool)
+    extends Module(_clock = clockIn, _reset = resetIn) {
+  val bits = log2Ceil(depth)
+  val io = new Bundle {
+    // These come from the sink domain
+    val deq  = Decoupled(gen)
+    // These cross to the source clock domain
+    val ridx = UInt(OUTPUT, width = bits+1)
+    val widx = UInt(INPUT,  width = bits+1)
+    val mem  = Vec(depth, gen).asInput
+  }
+
+  val ridx = GrayCounter(bits+1, io.deq.fire())
+  val widx = AsyncGrayCounter(io.widx, sync)
+  val valid = ridx =/= widx
+
+  // The mux is safe because timing analysis ensures ridx has reached the register
+  // On an ASIC, changes to the unread location cannot affect the selected value
+  // On an FPGA, only one input changes at a time => mem updates don't cause glitches
+  // The register only latches when the selected valued is not being written
+  val index = if (depth == 1) UInt(0) else ridx(bits-1, 0) ^ (ridx(bits, bits) << (bits-1))
+  io.deq.bits  := RegEnable(io.mem(index), valid && !reset)
+  io.deq.valid := RegNext(valid, Bool(false))
+  io.ridx := RegNext(ridx, UInt(0))
+}
+
+class AsyncQueue[T <: Data](gen: T, depth: Int = 8, sync: Int = 3) extends Crossing[T] {
+  require (sync >= 2)
+  require (depth > 0 && isPow2(depth))
+
+  val io = new CrossingIO(gen)
+  val source = Module(new AsyncQueueSource(gen, depth, sync, io.enq_clock, io.enq_reset))
+  val sink   = Module(new AsyncQueueSink  (gen, depth, sync, io.deq_clock, io.deq_reset))
+
+  source.io.enq <> io.enq
+  io.deq <> sink.io.deq
+
+  sink.io.mem := source.io.mem
+  sink.io.widx := source.io.widx
+  source.io.ridx := sink.io.ridx
+}

From fe6a67dd0eaac249fb1f40ddfb3a3b31a25a5e49 Mon Sep 17 00:00:00 2001
From: "Wesley W. Terpstra" <wesley@sifive.com>
Date: Tue, 13 Sep 2016 15:26:59 -0700
Subject: [PATCH 04/10] tilelink2: add a RegisterCrossing primitive

---
 .../scala/uncore/tilelink2/RegField.scala     |  13 ++
 .../uncore/tilelink2/RegisterCrossing.scala   | 130 ++++++++++++++++++
 2 files changed, 143 insertions(+)
 create mode 100644 src/main/scala/uncore/tilelink2/RegisterCrossing.scala

diff --git a/src/main/scala/uncore/tilelink2/RegField.scala b/src/main/scala/uncore/tilelink2/RegField.scala
index 1fac6a84..c3f758fc 100644
--- a/src/main/scala/uncore/tilelink2/RegField.scala
+++ b/src/main/scala/uncore/tilelink2/RegField.scala
@@ -17,6 +17,12 @@ object RegReadFn
   // effects must become visible on the cycle after ovalid && oready
   implicit def apply(x: (Bool, Bool) => (Bool, Bool, UInt)) =
     new RegReadFn(false, x)
+  implicit def apply(x: RegisterReadIO[UInt]): RegReadFn =
+    RegReadFn((ivalid, oready) => {
+      x.request.valid := ivalid
+      x.response.ready := oready
+      (x.request.ready, x.response.valid, x.response.bits)
+    })
   // (ready: Bool) => (valid: Bool, data: UInt)
   // valid must not combinationally depend on ready
   // valid must eventually go high without requiring ready to go high
@@ -47,6 +53,13 @@ object RegWriteFn
   // effects must become visible on the cycle after ovalid && oready
   implicit def apply(x: (Bool, Bool, UInt) => (Bool, Bool)) =
     new RegWriteFn(false, x)
+  implicit def apply(x: RegisterWriteIO[UInt]): RegWriteFn =
+    RegWriteFn((ivalid, oready, data) => {
+      x.request.valid := ivalid
+      x.request.bits := data
+      x.response.ready := oready
+      (x.request.ready, x.response.valid)
+    })
   // (valid: Bool, data: UInt) => (ready: Bool)
   // ready may combinationally depend on data (but not valid)
   // ready must eventually go high without requiring valid to go high
diff --git a/src/main/scala/uncore/tilelink2/RegisterCrossing.scala b/src/main/scala/uncore/tilelink2/RegisterCrossing.scala
new file mode 100644
index 00000000..36a5606e
--- /dev/null
+++ b/src/main/scala/uncore/tilelink2/RegisterCrossing.scala
@@ -0,0 +1,130 @@
+// See LICENSE for license details.
+
+package uncore.tilelink2
+
+import Chisel._
+import junctions._
+
+// A very simple flow control state machine, run in the specified clock domain
+class BusyRegisterCrossing(clock: Clock, reset: Bool)
+    extends Module(_clock = clock, _reset = reset) {
+  val io = new Bundle {
+    val progress       = Bool(INPUT)
+    val request_valid  = Bool(INPUT)
+    val response_ready = Bool(INPUT)
+    val busy           = Bool(OUTPUT)
+  }
+
+  val busy = RegInit(Bool(false))
+  when (io.progress) {
+    busy := Mux(busy, !io.response_ready, io.request_valid)
+  }
+  io.busy := busy
+}
+
+// RegField should support connecting to one of these
+class RegisterWriteIO[T <: Data](gen: T) extends Bundle {
+  val request  = Decoupled(gen).flip()
+  val response = Decoupled(Bool()) // ignore .bits
+}
+
+// To turn on/off a domain:
+//  1. lower allow on the other side
+//  2. wait for inflight traffic to resolve
+//  3. turn off the domain
+//  4. assert reset in the domain
+//  5. turn on the domain
+//  6. deassert reset in the domain
+//  7. raise allow on the other side
+
+class RegisterWriteCrossingIO[T <: Data](gen: T) extends Bundle {
+  // Master clock domain
+  val master_clock   = Clock(INPUT)
+  val master_reset   = Bool(INPUT)
+  val master_allow   = Bool(INPUT) // actually wait for the slave
+  val master_port    = new RegisterWriteIO(gen)
+  // Slave clock domain
+  val slave_clock    = Clock(INPUT)
+  val slave_reset    = Bool(INPUT)
+  val slave_allow    = Bool(INPUT) // honour requests from the master
+  val slave_register = gen.asOutput
+}
+
+class RegisterWriteCrossing[T <: Data](gen: T, sync: Int = 3) extends Module {
+  val io = new RegisterWriteCrossingIO(gen)
+  // The crossing must only allow one item inflight at a time
+  val crossing = Module(new AsyncQueue(gen, 1, sync))
+
+  // We can just randomly reset one-side of a single entry AsyncQueue.
+  // If the enq side is reset, at worst deq.bits is reassigned from mem(0), which stays fixed.
+  // If the deq side is reset, at worst the master rewrites mem(0) once, deq.bits stays fixed.
+  crossing.io.enq_clock := io.master_clock
+  crossing.io.enq_reset := io.master_reset || !io.master_allow
+  crossing.io.deq_clock := io.slave_clock
+  crossing.io.deq_reset := io.slave_reset || !io.slave_allow
+
+  crossing.io.enq.bits := io.master_port.request.bits
+  io.slave_register := crossing.io.deq.bits
+
+  // If the slave is not operational, just drop the write.
+  val progress = crossing.io.enq.ready || !io.master_allow
+
+  val reg = Module(new BusyRegisterCrossing(io.master_clock, io.master_reset))
+  reg.io.progress := progress
+  reg.io.request_valid  := io.master_port.request.valid
+  reg.io.response_ready := io.master_port.response.ready
+
+  crossing.io.deq.ready := Bool(true)
+  crossing.io.enq.valid := io.master_port.request.valid && !reg.io.busy
+  io.master_port.request.ready  := progress && !reg.io.busy
+  io.master_port.response.valid := progress && reg.io.busy
+}
+
+// RegField should support connecting to one of these
+class RegisterReadIO[T <: Data](gen: T) extends Bundle {
+  val request  = Decoupled(Bool()).flip() // ignore .bits
+  val response = Decoupled(gen)
+}
+
+class RegisterReadCrossingIO[T <: Data](gen: T) extends Bundle {
+  // Master clock domain
+  val master_clock   = Clock(INPUT)
+  val master_reset   = Bool(INPUT)
+  val master_allow   = Bool(INPUT) // actually wait for the slave
+  val master_port    = new RegisterReadIO(gen)
+  // Slave clock domain
+  val slave_clock    = Clock(INPUT)
+  val slave_reset    = Bool(INPUT)
+  val slave_allow    = Bool(INPUT) // honour requests from the master
+  val slave_register = gen.asInput
+}
+
+class RegisterReadCrossing[T <: Data](gen: T, sync: Int = 3) extends Module {
+  val io = new RegisterReadCrossingIO(gen)
+  // The crossing must only allow one item inflight at a time
+  val crossing = Module(new AsyncQueue(gen, 1, sync))
+
+  // We can just randomly reset one-side of a single entry AsyncQueue.
+  // If the enq side is reset, at worst deq.bits is reassigned from mem(0), which stays fixed.
+  // If the deq side is reset, at worst the slave rewrites mem(0) once, deq.bits stays fixed.
+  crossing.io.enq_clock := io.slave_clock
+  crossing.io.enq_reset := io.slave_reset || !io.slave_allow
+  crossing.io.deq_clock := io.master_clock
+  crossing.io.deq_reset := io.master_reset || !io.master_allow
+
+  crossing.io.enq.bits := io.slave_register
+  io.master_port.response.bits := crossing.io.deq.bits
+
+  // If the slave is not operational, just repeat the last value we saw.
+  val progress = crossing.io.deq.valid || !io.master_allow
+
+  val reg = Module(new BusyRegisterCrossing(io.master_clock, io.master_reset))
+  reg.io.progress := progress
+  reg.io.request_valid  := io.master_port.request.valid
+  reg.io.response_ready := io.master_port.response.ready
+
+  io.master_port.response.valid := progress && reg.io.busy
+  io.master_port.request.ready  := progress && !reg.io.busy
+  crossing.io.deq.ready := io.master_port.request.valid && !reg.io.busy
+  crossing.io.enq.valid := Bool(true)
+}

From 3348236320a4576a35ee8666675d9fc260a60d06 Mon Sep 17 00:00:00 2001
From: "Wesley W. Terpstra" <wesley@sifive.com>
Date: Tue, 13 Sep 2016 15:45:52 -0700
Subject: [PATCH 05/10] junctions: remove obsolete Handshaker crossing

---
 src/main/scala/junctions/crossing.scala | 100 +-----------------------
 1 file changed, 4 insertions(+), 96 deletions(-)

diff --git a/src/main/scala/junctions/crossing.scala b/src/main/scala/junctions/crossing.scala
index f0db6f86..948c3a21 100644
--- a/src/main/scala/junctions/crossing.scala
+++ b/src/main/scala/junctions/crossing.scala
@@ -16,106 +16,14 @@ abstract class Crossing[T <: Data] extends Module {
   val io: CrossingIO[T]
 }
 
-// Output is 1 for one cycle after any edge of 'in'
-object AsyncHandshakePulse {
-  def apply(in: Bool, sync: Int): Bool = {
-    val syncv = RegInit(Vec.fill(sync+1){Bool(false)})
-    syncv.last := in
-    (syncv.init zip syncv.tail).foreach { case (sink, source) => sink := source }
-    syncv(0) =/= syncv(1)
-  }
-}
-
-class AsyncHandshakeSource[T <: Data](gen: T, sync: Int, clock: Clock, reset: Bool)
-    extends Module(_clock = clock, _reset = reset) {
-  val io = new Bundle {
-    // These come from the source clock domain
-    val enq  = Decoupled(gen).flip()
-    // These cross to the sink clock domain
-    val bits = gen.cloneType.asOutput
-    val push = Bool(OUTPUT)
-    val pop  = Bool(INPUT)
-  }
-
-  val ready = RegInit(Bool(true))
-  val bits = Reg(gen)
-  val push = RegInit(Bool(false))
-
-  io.enq.ready := ready
-  io.bits := bits
-  io.push := push
-
-  val pop = AsyncHandshakePulse(io.pop, sync)
-  assert (!pop || !ready)
-
-  when (pop) {
-    ready := Bool(true)
-  }
-
-  when (io.enq.fire()) {
-    ready := Bool(false)
-    bits := io.enq.bits
-    push := !push
-  }
-}
-
-class AsyncHandshakeSink[T <: Data](gen: T, sync: Int, clock: Clock, reset: Bool) 
-    extends Module(_clock = clock, _reset = reset) {
-  val io = new Bundle {
-    // These cross to the source clock domain
-    val bits = gen.cloneType.asInput
-    val push = Bool(INPUT)
-    val pop  = Bool(OUTPUT)
-    // These go to the sink clock domain
-    val deq = Decoupled(gen)
-  }
-
-  val valid = RegInit(Bool(false))
-  val bits  = Reg(gen)
-  val pop   = RegInit(Bool(false))
-
-  io.deq.valid := valid
-  io.deq.bits  := bits
-  io.pop := pop
-
-  val push = AsyncHandshakePulse(io.push, sync)
-  assert (!push || !valid)
-
-  when (push) {
-    valid := Bool(true)
-    bits  := io.bits
-  }
-
-  when (io.deq.fire()) {
-    valid := Bool(false)
-    pop := !pop
-  }
-}
-
-class AsyncHandshake[T <: Data](gen: T, sync: Int = 2) extends Crossing[T] {
-  val io = new CrossingIO(gen)
-  require (sync >= 2)
-
-  val source = Module(new AsyncHandshakeSource(gen, sync, io.enq_clock, io.enq_reset))
-  val sink   = Module(new AsyncHandshakeSink  (gen, sync, io.deq_clock, io.deq_reset))
-
-  source.io.enq <> io.enq
-  io.deq <> sink.io.deq
-
-  sink.io.bits := source.io.bits
-  sink.io.push := source.io.push
-  source.io.pop := sink.io.pop
-}
-
 class AsyncScope extends Module { val io = new Bundle }
 object AsyncScope { def apply() = Module(new AsyncScope) }
 
 object AsyncDecoupledCrossing
 {
   // takes from_source from the 'from' clock domain and puts it into the 'to' clock domain
-  def apply[T <: Data](from_clock: Clock, from_reset: Bool, from_source: DecoupledIO[T], to_clock: Clock, to_reset: Bool, depth: Int = 3, sync: Int = 2): DecoupledIO[T] = {
-    // !!! if depth == 0 { use Handshake } else { use AsyncFIFO }
-    val crossing = Module(new AsyncHandshake(from_source.bits, sync)).io
+  def apply[T <: Data](from_clock: Clock, from_reset: Bool, from_source: DecoupledIO[T], to_clock: Clock, to_reset: Bool, depth: Int = 8, sync: Int = 3): DecoupledIO[T] = {
+    val crossing = Module(new AsyncQueue(from_source.bits, depth, sync)).io
     crossing.enq_clock := from_clock
     crossing.enq_reset := from_reset
     crossing.enq       <> from_source
@@ -128,7 +36,7 @@ object AsyncDecoupledCrossing
 object AsyncDecoupledTo
 {
   // takes source from your clock domain and puts it into the 'to' clock domain
-  def apply[T <: Data](to_clock: Clock, to_reset: Bool, source: DecoupledIO[T], depth: Int = 3, sync: Int = 2): DecoupledIO[T] = {
+  def apply[T <: Data](to_clock: Clock, to_reset: Bool, source: DecoupledIO[T], depth: Int = 8, sync: Int = 3): DecoupledIO[T] = {
     val scope = AsyncScope()
     AsyncDecoupledCrossing(scope.clock, scope.reset, source, to_clock, to_reset, depth, sync)
   }
@@ -137,7 +45,7 @@ object AsyncDecoupledTo
 object AsyncDecoupledFrom
 {
   // takes from_source from the 'from' clock domain and puts it into your clock domain
-  def apply[T <: Data](from_clock: Clock, from_reset: Bool, from_source: DecoupledIO[T], depth: Int = 3, sync: Int = 2): DecoupledIO[T] = {
+  def apply[T <: Data](from_clock: Clock, from_reset: Bool, from_source: DecoupledIO[T], depth: Int = 8, sync: Int = 3): DecoupledIO[T] = {
     val scope = AsyncScope()
     AsyncDecoupledCrossing(from_clock, from_reset, from_source, scope.clock, scope.reset, depth, sync)
   }

From 44501cdbf80f2f159d41441d3a6e9f4a59e52344 Mon Sep 17 00:00:00 2001
From: "Wesley W. Terpstra" <wesley@sifive.com>
Date: Tue, 13 Sep 2016 15:49:08 -0700
Subject: [PATCH 06/10] crossings: change defaults to sync=3 for safer settling
 time

Make the matching AsyncQueue depth=8 to support full throughput
---
 src/main/scala/junctions/nasti.scala      | 6 +++---
 src/main/scala/uncore/devices/Debug.scala | 4 ++--
 2 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/src/main/scala/junctions/nasti.scala b/src/main/scala/junctions/nasti.scala
index 3ae1fbc2..43803153 100644
--- a/src/main/scala/junctions/nasti.scala
+++ b/src/main/scala/junctions/nasti.scala
@@ -708,7 +708,7 @@ class NastiMemoryDemux(nRoutes: Int)(implicit p: Parameters) extends NastiModule
 
 object AsyncNastiCrossing {
   // takes from_source from the 'from' clock domain to the 'to' clock domain
-  def apply(from_clock: Clock, from_reset: Bool, from_source: NastiIO, to_clock: Clock, to_reset: Bool, depth: Int = 3, sync: Int = 2) = {
+  def apply(from_clock: Clock, from_reset: Bool, from_source: NastiIO, to_clock: Clock, to_reset: Bool, depth: Int = 8, sync: Int = 3) = {
     val to_sink = Wire(new NastiIO()(from_source.p))
 
     to_sink.aw <> AsyncDecoupledCrossing(from_clock, from_reset, from_source.aw, to_clock, to_reset, depth, sync)
@@ -723,7 +723,7 @@ object AsyncNastiCrossing {
 
 object AsyncNastiTo {
   // takes source from your clock domain and puts it into the 'to' clock domain
-  def apply(to_clock: Clock, to_reset: Bool, source: NastiIO, depth: Int = 3, sync: Int = 2): NastiIO = {
+  def apply(to_clock: Clock, to_reset: Bool, source: NastiIO, depth: Int = 8, sync: Int = 3): NastiIO = {
     val scope = AsyncScope()
     AsyncNastiCrossing(scope.clock, scope.reset, source, to_clock, to_reset, depth, sync)
   }
@@ -731,7 +731,7 @@ object AsyncNastiTo {
 
 object AsyncNastiFrom {
   // takes from_source from the 'from' clock domain and puts it into your clock domain
-  def apply(from_clock: Clock, from_reset: Bool, from_source: NastiIO, depth: Int = 3, sync: Int = 2): NastiIO = {
+  def apply(from_clock: Clock, from_reset: Bool, from_source: NastiIO, depth: Int = 8, sync: Int = 3): NastiIO = {
     val scope = AsyncScope()
     AsyncNastiCrossing(from_clock, from_reset, from_source, scope.clock, scope.reset, depth, sync)
   }
diff --git a/src/main/scala/uncore/devices/Debug.scala b/src/main/scala/uncore/devices/Debug.scala
index 310bdc68..6714a741 100644
--- a/src/main/scala/uncore/devices/Debug.scala
+++ b/src/main/scala/uncore/devices/Debug.scala
@@ -994,7 +994,7 @@ object AsyncDebugBusCrossing {
 
 object AsyncDebugBusFrom { // OutsideClockDomain
   // takes from_source from the 'from' clock domain and puts it into your clock domain
-  def apply(from_clock: Clock, from_reset: Bool, from_source: DebugBusIO, depth: Int = 0, sync: Int = 2): DebugBusIO = {
+  def apply(from_clock: Clock, from_reset: Bool, from_source: DebugBusIO, depth: Int = 1, sync: Int = 3): DebugBusIO = {
     val scope = AsyncScope()
     AsyncDebugBusCrossing(from_clock, from_reset, from_source, scope.clock, scope.reset, depth, sync)
   }
@@ -1002,7 +1002,7 @@ object AsyncDebugBusFrom { // OutsideClockDomain
 
 object AsyncDebugBusTo { // OutsideClockDomain
   // takes source from your clock domain and puts it into the 'to' clock domain
-  def apply(to_clock: Clock, to_reset: Bool, source: DebugBusIO, depth: Int = 0, sync: Int = 2): DebugBusIO = {
+  def apply(to_clock: Clock, to_reset: Bool, source: DebugBusIO, depth: Int = 1, sync: Int = 3): DebugBusIO = {
     val scope = AsyncScope()
     AsyncDebugBusCrossing(scope.clock, scope.reset, source, to_clock, to_reset, depth, sync)
   }

From c8e6d478840909915ac69aa38820b132d9dd50c7 Mon Sep 17 00:00:00 2001
From: "Wesley W. Terpstra" <wesley@sifive.com>
Date: Tue, 13 Sep 2016 16:04:46 -0700
Subject: [PATCH 07/10] tilelink2: add a clock crossing adapter

---
 src/main/scala/uncore/tilelink2/Buffer.scala  |  2 +-
 .../scala/uncore/tilelink2/Crossing.scala     | 42 +++++++++++++++++++
 2 files changed, 43 insertions(+), 1 deletion(-)
 create mode 100644 src/main/scala/uncore/tilelink2/Crossing.scala

diff --git a/src/main/scala/uncore/tilelink2/Buffer.scala b/src/main/scala/uncore/tilelink2/Buffer.scala
index adfe4c4b..558a524a 100644
--- a/src/main/scala/uncore/tilelink2/Buffer.scala
+++ b/src/main/scala/uncore/tilelink2/Buffer.scala
@@ -22,7 +22,7 @@ class TLBuffer(entries: Int = 2, pipe: Boolean = false) extends LazyModule
       if (edgeOut.manager.anySupportAcquire && edgeOut.client.anySupportProbe) {
         in .b <> Queue(out.b, entries, pipe)
         out.c <> Queue(in .c, entries, pipe)
-        out.e <> Queue(out.e, entries, pipe)
+        out.e <> Queue(in .e, entries, pipe)
       } else {
         in.b.valid := Bool(false)
         in.c.ready := Bool(true)
diff --git a/src/main/scala/uncore/tilelink2/Crossing.scala b/src/main/scala/uncore/tilelink2/Crossing.scala
new file mode 100644
index 00000000..1f903c9a
--- /dev/null
+++ b/src/main/scala/uncore/tilelink2/Crossing.scala
@@ -0,0 +1,42 @@
+// See LICENSE for license details.
+
+package uncore.tilelink2
+
+import Chisel._
+import chisel3.internal.sourceinfo.SourceInfo
+import junctions._
+
+class TLAsyncCrossing(depth: Int = 8, sync: Int = 3) extends LazyModule
+{
+  val node = TLIdentityNode()
+
+  lazy val module = new LazyModuleImp(this) {
+    val io = new Bundle {
+      val in        = node.bundleIn
+      val in_clock  = Clock(INPUT)
+      val in_reset  = Bool(INPUT)
+      val out       = node.bundleOut
+      val out_clock = Clock(INPUT)
+      val out_reset = Bool(INPUT)
+    }
+
+    // Transfer all TL2 bundles from/to the same domains
+    ((io.in zip io.out) zip (node.edgesIn zip node.edgesOut)) foreach { case ((in, out), (edgeIn, edgeOut)) =>
+      out.a <> AsyncDecoupledCrossing(io.in_clock, io.in_reset, in.a, io.out_clock, io.out_reset, depth, sync)
+      in.d <> AsyncDecoupledCrossing(io.out_clock, io.out_reset, out.d, io.in_clock, io.in_reset, depth, sync)
+
+      if (edgeOut.manager.anySupportAcquire && edgeOut.client.anySupportProbe) {
+        in.b <> AsyncDecoupledCrossing(io.out_clock, io.out_reset, out.b, io.in_clock, io.in_reset, depth, sync)
+        out.c <> AsyncDecoupledCrossing(io.in_clock, io.in_reset, in.c, io.out_clock, io.out_reset, depth, sync)
+        out.e <> AsyncDecoupledCrossing(io.in_clock, io.in_reset, in.e, io.out_clock, io.out_reset, depth, sync)
+      } else {
+        in.b.valid := Bool(false)
+        in.c.ready := Bool(true)
+        in.e.ready := Bool(true)
+        out.b.ready := Bool(true)
+        out.c.valid := Bool(false)
+        out.e.valid := Bool(false)
+      }
+    }
+  }
+}

From acedd3688a298b6022c597d83773fbe3813a8a52 Mon Sep 17 00:00:00 2001
From: "Wesley W. Terpstra" <wesley@sifive.com>
Date: Tue, 13 Sep 2016 16:35:06 -0700
Subject: [PATCH 08/10] tilelink2: unit test for the clock crossing

---
 src/main/scala/uncore/tilelink2/Fuzzer.scala | 26 +++++++++++++++++++-
 vsrc/ClockDivider.v                          | 19 ++++++++++++++
 2 files changed, 44 insertions(+), 1 deletion(-)
 create mode 100644 vsrc/ClockDivider.v

diff --git a/src/main/scala/uncore/tilelink2/Fuzzer.scala b/src/main/scala/uncore/tilelink2/Fuzzer.scala
index b2a734f8..b5cdd263 100644
--- a/src/main/scala/uncore/tilelink2/Fuzzer.scala
+++ b/src/main/scala/uncore/tilelink2/Fuzzer.scala
@@ -206,6 +206,15 @@ class TLFuzzer(
   }
 }
 
+class ClockDivider extends BlackBox {
+  val io = new Bundle {
+    val clock_in  = Clock(INPUT)
+    val reset_in  = Bool(INPUT)
+    val clock_out = Clock(OUTPUT)
+    val reset_out = Bool(OUTPUT)
+  }
+}
+
 class TLFuzzRAM extends LazyModule
 {
   val model = LazyModule(new TLRAMModel)
@@ -213,14 +222,29 @@ class TLFuzzRAM extends LazyModule
   val gpio = LazyModule(new RRTest1(0x400))
   val xbar = LazyModule(new TLXbar)
   val fuzz = LazyModule(new TLFuzzer(5000))
+  val cross = LazyModule(new TLAsyncCrossing)
 
   model.node := fuzz.node
   xbar.node := TLWidthWidget(TLHintHandler(model.node), 16)
-  ram.node := TLFragmenter(TLBuffer(xbar.node), 4, 256)
+  cross.node := TLFragmenter(TLBuffer(xbar.node), 4, 256)
+  ram.node := cross.node
   gpio.node := TLFragmenter(TLBuffer(xbar.node), 4, 32)
 
   lazy val module = new LazyModuleImp(this) with HasUnitTestIO {
     io.finished := fuzz.module.io.finished
+
+    // Shove the RAM into another clock domain
+    val clocks = Module(new ClockDivider)
+    ram.module.clock := clocks.io.clock_out
+    ram.module.reset := clocks.io.reset_out
+    clocks.io.clock_in := clock
+    clocks.io.reset_in := reset
+
+    // ... and safely cross TL2 into it
+    cross.module.io.in_clock := clock
+    cross.module.io.in_reset := reset
+    cross.module.io.out_clock := clocks.io.clock_out
+    cross.module.io.out_reset := clocks.io.reset_out
   }
 }
 
diff --git a/vsrc/ClockDivider.v b/vsrc/ClockDivider.v
new file mode 100644
index 00000000..83bb561b
--- /dev/null
+++ b/vsrc/ClockDivider.v
@@ -0,0 +1,19 @@
+// You can't divide clocks in Chisel
+module ClockDivider(
+  input  clock_in,
+  input  reset_in,
+  output clock_out,
+  output reset_out
+);
+
+  reg [2:0] shift = 3'b001;
+
+  always @(posedge clock_in)
+  begin
+    shift <= {shift[0], shift[2:1]};
+  end
+  
+  assign reset_out = reset_in;
+  assign clock_out = shift[0];
+
+endmodule

From cc88bf1b089156697b924757e8c4f4e8ab713e46 Mon Sep 17 00:00:00 2001
From: "Wesley W. Terpstra" <wesley@sifive.com>
Date: Tue, 13 Sep 2016 17:41:35 -0700
Subject: [PATCH 09/10] junctions: give unit tests more time

---
 src/main/scala/junctions/unittests/UnitTest.scala | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/main/scala/junctions/unittests/UnitTest.scala b/src/main/scala/junctions/unittests/UnitTest.scala
index 77dd146b..b64e0f5f 100644
--- a/src/main/scala/junctions/unittests/UnitTest.scala
+++ b/src/main/scala/junctions/unittests/UnitTest.scala
@@ -39,7 +39,7 @@ class UnitTestSuite(implicit p: Parameters) extends Module {
     state := Mux(test_idx === UInt(tests.size - 1), s_done, s_start)
   }
 
-  val timer = Module(new Timer(100000, tests.size))
+  val timer = Module(new Timer(500000, tests.size))
   timer.io.start.valid := Bool(false)
   timer.io.stop.valid := Bool(false)
 

From d23ab7370dd172db9925b466561f14a4d82eae5f Mon Sep 17 00:00:00 2001
From: "Wesley W. Terpstra" <wesley@sifive.com>
Date: Tue, 13 Sep 2016 18:33:29 -0700
Subject: [PATCH 10/10] tilelink2: Unit Test for the RegisterCrossing

---
 .../uncore/tilelink2/RegisterRouterTest.scala | 32 +++++++++++++++++--
 1 file changed, 30 insertions(+), 2 deletions(-)

diff --git a/src/main/scala/uncore/tilelink2/RegisterRouterTest.scala b/src/main/scala/uncore/tilelink2/RegisterRouterTest.scala
index 65dde1d3..be81b4c8 100644
--- a/src/main/scala/uncore/tilelink2/RegisterRouterTest.scala
+++ b/src/main/scala/uncore/tilelink2/RegisterRouterTest.scala
@@ -212,9 +212,37 @@ trait RRTest1Bundle
 {
 }
 
-trait RRTest1Module extends HasRegMap
+trait RRTest1Module extends Module with HasRegMap
 {
-  regmap(RRTest1Map.map:_*)
+  val clocks = Module(new ClockDivider)
+  clocks.io.clock_in := clock
+  clocks.io.reset_in := reset
+
+  def x(bits: Int) = {
+    val field = UInt(width = bits)
+
+    val readCross = Module(new RegisterReadCrossing(field))
+    readCross.io.master_clock := clock
+    readCross.io.master_reset := reset
+    readCross.io.master_allow := Bool(true)
+    readCross.io.slave_clock := clocks.io.clock_out
+    readCross.io.slave_reset := clocks.io.reset_out
+    readCross.io.slave_allow := Bool(true)
+
+    val writeCross = Module(new RegisterWriteCrossing(field))
+    writeCross.io.master_clock := clock
+    writeCross.io.master_reset := reset
+    writeCross.io.master_allow := Bool(true)
+    writeCross.io.slave_clock := clocks.io.clock_out
+    writeCross.io.slave_reset := clocks.io.reset_out
+    writeCross.io.slave_allow := Bool(true)
+
+    readCross.io.slave_register := writeCross.io.slave_register
+    RegField(bits, readCross.io.master_port, writeCross.io.master_port)
+  }
+
+  val map = RRTest1Map.map.drop(1) ++ Seq(0 -> Seq(x(8), x(8), x(8), x(8)))
+  regmap(map:_*)
 }
 
 class RRTest1(address: BigInt) extends TLRegisterRouter(address, 0, 32, Some(6), 4)(