Merge pull request #1489 from ucb-bar/clkfixes

Improvements to chipyard clocking

.github/workflows/chipyard-full-flow.yml

@@ -80,7 +80,7 @@ jobs:
           eval "$(conda shell.bash hook)"
           mkdir ${{ env.JAVA_TMP_DIR }}
           export MAKEFLAGS="-j32"
-          ./build-setup.sh -f
+          ./build-setup.sh -f -v
 
   run-cfg-finder:
     name: run-cfg-finder

generators/chipyard/src/main/scala/clocking/ClockBinders.scala

@@ -38,9 +38,9 @@ class WithPLLSelectorDividerClockGenerator extends OverrideLazyIOBinder({
     val clockSelector = system.prci_ctrl_domain { LazyModule(new TLClockSelector(baseAddress + 0x30000, tlbus.beatBytes)) }
     val pllCtrl       = system.prci_ctrl_domain { LazyModule(new FakePLLCtrl    (baseAddress + 0x40000, tlbus.beatBytes)) }
 
-    tlbus.coupleTo("clock-div-ctrl") { clockDivider.tlNode := TLFragmenter(tlbus.beatBytes, tlbus.blockBytes) := TLBuffer() := _ }
-    tlbus.coupleTo("clock-sel-ctrl") { clockSelector.tlNode := TLFragmenter(tlbus.beatBytes, tlbus.blockBytes) := TLBuffer() := _ }
-    tlbus.coupleTo("pll-ctrl") { pllCtrl.tlNode := TLFragmenter(tlbus.beatBytes, tlbus.blockBytes) := TLBuffer() := _ }
+    clockDivider.tlNode  := system.prci_ctrl_bus
+    clockSelector.tlNode := system.prci_ctrl_bus
+    pllCtrl.tlNode       := system.prci_ctrl_bus
 
     system.allClockGroupsNode := clockDivider.clockNode := clockSelector.clockNode
 

generators/chipyard/src/main/scala/clocking/HasChipyardPRCI.scala

@@ -36,6 +36,14 @@ trait HasChipyardPRCI { this: BaseSubsystem with InstantiatesTiles =>
   val prci_ctrl_domain = LazyModule(new ClockSinkDomain(name=Some("chipyard-prci-control")))
   prci_ctrl_domain.clockNode := tlbus.fixedClockNode
 
+  val prci_ctrl_bus = prci_ctrl_domain { TLXbar() }
+  tlbus.coupleTo("prci_ctrl") { (prci_ctrl_bus
+    := TLFIFOFixer(TLFIFOFixer.all)
+    := TLFragmenter(tlbus.beatBytes, tlbus.blockBytes)
+    := TLBuffer()
+    := _)
+  }
+
   // Aggregate all the clock groups into a single node
   val aggregator = LazyModule(new ClockGroupAggregator("allClocks")).node
   val allClockGroupsNode = ClockGroupEphemeralNode()
@@ -71,19 +79,24 @@ trait HasChipyardPRCI { this: BaseSubsystem with InstantiatesTiles =>
   // diplomatic IOBinder should drive
   val frequencySpecifier = ClockGroupFrequencySpecifier(p(ClockFrequencyAssignersKey))
   val clockGroupCombiner = ClockGroupCombiner()
-  val resetSynchronizer  = ClockGroupResetSynchronizer()
-  val tileClockGater     = if (prciParams.enableTileClockGating) { prci_ctrl_domain {
-    TileClockGater(prciParams.baseAddress + 0x00000, tlbus)
-  } } else { ClockGroupEphemeralNode() }
-  val tileResetSetter    = if (prciParams.enableTileResetSetting) { prci_ctrl_domain {
-    TileResetSetter(prciParams.baseAddress + 0x10000, tlbus, tile_prci_domains.map(_.tile_reset_domain.clockNode.portParams(0).name.get), Nil)
-  } } else { ClockGroupEphemeralNode() }
+  val resetSynchronizer  = prci_ctrl_domain { ClockGroupResetSynchronizer() }
+  val tileClockGater     = Option.when(prciParams.enableTileClockGating) { prci_ctrl_domain {
+    val clock_gater = LazyModule(new TileClockGater(prciParams.baseAddress + 0x00000, tlbus.beatBytes))
+    clock_gater.tlNode := prci_ctrl_bus
+    clock_gater
+  } }
+  val tileResetSetter    = Option.when(prciParams.enableTileResetSetting) { prci_ctrl_domain {
+    val reset_setter = LazyModule(new TileResetSetter(prciParams.baseAddress + 0x10000, tlbus.beatBytes,
+      tile_prci_domains.map(_.tile_reset_domain.clockNode.portParams(0).name.get), Nil))
+    reset_setter.tlNode := prci_ctrl_bus
+    reset_setter
+  } }
 
   (aggregator
     := frequencySpecifier
     := clockGroupCombiner
     := resetSynchronizer
-    := tileClockGater
-    := tileResetSetter
+    := tileClockGater.map(_.clockNode).getOrElse(ClockGroupEphemeralNode()(ValName("temp")))
+    := tileResetSetter.map(_.clockNode).getOrElse(ClockGroupEphemeralNode()(ValName("temp")))
     := allClockGroupsNode)
 }

generators/chipyard/src/main/scala/clocking/TLClockDivider.scala

@@ -26,20 +26,26 @@ class TLClockDivider(address: BigInt, beatBytes: Int, divBits: Int = 8)(implicit
     val sinks = clockNode.out.head._1.member.elements.toSeq
     require (sources.size == sinks.size)
     val nSinks = sinks.size
+    // The implicit clock of this module is the clock of the tilelink bus
+    // busReset is sync'd to that clock, and will be asserted longer than the
+    // resets coming in through the clockNode, since the busReset is derived from
+    // the clockNode resets in downstream PRCI nodes
+    val busReset = reset
 
     val regs = (0 until nSinks) .map { i =>
       val sinkName = sinks(i)._1
       val asyncReset = sources(i).reset
-      val reg = withReset (asyncReset) {
-        Module(new AsyncResetRegVec(w=divBits, init=0))
-      }
+      val reg = Module(new AsyncResetRegVec(w=divBits, init=0))
+
       println(s"${(address+i*4).toString(16)}: Clock domain $sinkName divider")
-      sinks(i)._2.clock := withClockAndReset(sources(i).clock, asyncReset) {
-        val divider = Module(new testchipip.ClockDivideOrPass(divBits, depth = 3, genClockGate = p(ClockGateImpl)))
-        divider.io.divisor := reg.io.q
-        divider.io.resetAsync := ResetStretcher(sources(i).clock, asyncReset, 20).asAsyncReset
-        divider.io.clockOut
-      }
+      val divider = Module(new testchipip.ClockDivideOrPass(divBits, depth = 3, genClockGate = p(ClockGateImpl)))
+      divider.io.clockIn := sources(i).clock
+      // busReset is expected to be high for a long time, since reset will take a while to propagate
+      // to the TL bus. While reset is propagating, make sure we propagate a fast, undivided clock
+      // by setting divisor=0. The divisor signal into the ClockDividerOrPass is synchronized internally
+      divider.io.divisor := Mux(busReset.asBool, 0.U, reg.io.q)
+      divider.io.resetAsync := ResetStretcher(sources(i).clock, asyncReset, 20).asAsyncReset
+      sinks(i)._2.clock := divider.io.clockOut
 
       // Note this is not synchronized to the output clock, which takes time to appear
       // so this is still asyncreset

generators/chipyard/src/main/scala/clocking/TileClockGater.scala

@@ -46,10 +46,3 @@ class TileClockGater(address: BigInt, beatBytes: Int)(implicit p: Parameters, va
   }
 }
 
-object TileClockGater {
-  def apply(address: BigInt, tlbus: TLBusWrapper)(implicit p: Parameters, v: ValName) = {
-    val gater = LazyModule(new TileClockGater(address, tlbus.beatBytes))
-    tlbus.coupleTo("clock-gater") { gater.tlNode := TLFragmenter(tlbus.beatBytes, tlbus.blockBytes) := TLBuffer() := _ }
-    gater.clockNode
-  }
-}

generators/chipyard/src/main/scala/clocking/TileResetSetter.scala

@@ -62,12 +62,3 @@ class TileResetSetter(address: BigInt, beatBytes: Int, tileNames: Seq[String], i
     }
   }
 }
-
-
-object TileResetSetter {
-  def apply(address: BigInt, tlbus: TLBusWrapper, tileNames: Seq[String], initResetHarts: Seq[Int])(implicit p: Parameters, v: ValName) = {
-    val setter = LazyModule(new TileResetSetter(address, tlbus.beatBytes, tileNames, initResetHarts))
-    tlbus.coupleTo("tile-reset-setter") { setter.tlNode := TLFragmenter(tlbus.beatBytes, tlbus.blockBytes) := TLBuffer() := _ }
-    setter.clockNode
-  }
-}