diff --git a/src/main/scala/coreplex/Configs.scala b/src/main/scala/coreplex/Configs.scala
index 44dbcadd..c320390c 100644
--- a/src/main/scala/coreplex/Configs.scala
+++ b/src/main/scala/coreplex/Configs.scala
@@ -151,10 +151,12 @@ class WithBufferlessBroadcastHub extends Config((site, here, up) => {
  * DO NOT use this configuration.
  */
 class WithStatelessBridge extends Config((site, here, up) => {
-  case BankedL2Key => up(BankedL2Key, site).copy(coherenceManager = { case (q, _) =>
-    implicit val p = q
+  case BankedL2Key => up(BankedL2Key, site).copy(coherenceManager = { coreplex =>
+    implicit val p = coreplex.p
     val cork = LazyModule(new TLCacheCork(unsafe = true))
-    (cork.node, cork.node)
+    val ww = LazyModule(new TLWidthWidget(coreplex.sbusBeatBytes))
+    ww.node :*= cork.node
+    (cork.node, ww.node, () => None)
   })
 })
 
diff --git a/src/main/scala/coreplex/MemoryBus.scala b/src/main/scala/coreplex/MemoryBus.scala
index 879a08e9..2a2f5286 100644
--- a/src/main/scala/coreplex/MemoryBus.scala
+++ b/src/main/scala/coreplex/MemoryBus.scala
@@ -22,12 +22,13 @@ case object BroadcastKey extends Field(BroadcastParams())
 case class BankedL2Params(
   nMemoryChannels:  Int = 1,
   nBanksPerChannel: Int = 1,
-  coherenceManager: (Parameters, HasMemoryBus) => (TLInwardNode, TLOutwardNode) = { case (q, _) =>
-    implicit val p = q
-    val MemoryBusParams(_, blockBytes, _, _) = p(MemoryBusKey)
+  coherenceManager: HasMemoryBus => (TLInwardNode, TLOutwardNode, () => Option[Bool]) = { coreplex =>
+    implicit val p = coreplex.p
     val BroadcastParams(nTrackers, bufferless) = p(BroadcastKey)
-    val bh = LazyModule(new TLBroadcast(blockBytes, nTrackers, bufferless))
-    (bh.node, bh.node)
+    val bh = LazyModule(new TLBroadcast(coreplex.memBusBlockBytes, nTrackers, bufferless))
+    val ww = LazyModule(new TLWidthWidget(coreplex.sbusBeatBytes))
+    ww.node :*= bh.node
+    (bh.node, ww.node, () => None)
   }) {
   val nBanks = nMemoryChannels*nBanksPerChannel
 }
@@ -53,24 +54,25 @@ class MemoryBus(params: MemoryBusParams)(implicit p: Parameters) extends TLBusWr
 
 trait HasMemoryBus extends HasSystemBus with HasPeripheryBus with HasInterruptBus {
   private val mbusParams = p(MemoryBusKey)
-  private val MemoryBusParams(beatBytes, blockBytes, _, _) = mbusParams
   private val l2Params = p(BankedL2Key)
+  val MemoryBusParams(memBusBeatBytes, memBusBlockBytes, _, _) = mbusParams
   val BankedL2Params(nMemoryChannels, nBanksPerChannel, coherenceManager) = l2Params
   val nBanks = l2Params.nBanks
-  val cacheBlockBytes = blockBytes
+  val cacheBlockBytes = memBusBlockBytes
+  private val (in, out, halt) = coherenceManager(this)
+  def memBusCanCauseHalt: () => Option[Bool] = halt
 
   require (isPow2(nMemoryChannels) || nMemoryChannels == 0)
   require (isPow2(nBanksPerChannel))
-  require (isPow2(blockBytes))
+  require (isPow2(memBusBlockBytes))
 
-  private val (in, out) = coherenceManager(p, this)
-  private val mask = ~BigInt((nBanks-1) * blockBytes)
+  private val mask = ~BigInt((nBanks-1) * memBusBlockBytes)
   val memBuses = Seq.tabulate(nMemoryChannels) { channel =>
     val mbus = LazyModule(new MemoryBus(mbusParams))
     for (bank <- 0 until nBanksPerChannel) {
       val offset = (bank * nMemoryChannels) + channel
       ForceFanout(a = true) { implicit p => in := sbus.toMemoryBus }
-      mbus.fromCoherenceManager := TLFilter(TLFilter.Mmask(AddressSet(offset * blockBytes, mask)))(out)
+      mbus.fromCoherenceManager := TLFilter(TLFilter.Mmask(AddressSet(offset * memBusBlockBytes, mask)))(out)
     }
     mbus
   }
diff --git a/src/main/scala/rocket/CSR.scala b/src/main/scala/rocket/CSR.scala
index 8051946b..c3a05712 100644
--- a/src/main/scala/rocket/CSR.scala
+++ b/src/main/scala/rocket/CSR.scala
@@ -155,7 +155,8 @@ class TracedInstruction(implicit p: Parameters) extends CoreBundle {
   val insn = UInt(width = iLen)
   val priv = UInt(width = 3)
   val exception = Bool()
-  val cause = UInt(width = 1 + log2Ceil(xLen))
+  val interrupt = Bool()
+  val cause = UInt(width = log2Ceil(xLen))
   val tval = UInt(width = coreMaxAddrBits max iLen)
 }
 
@@ -519,7 +520,7 @@ class CSRFile(perfEventSets: EventSets = new EventSets(Seq()))(implicit p: Param
   assert(!reg_singleStepped || io.retire === UInt(0))
 
   val epc = ~(~io.pc | (coreInstBytes-1))
-  val write_badaddr = cause isOneOf (Causes.illegal_instruction, Causes.breakpoint,
+  val write_badaddr = exception && cause.isOneOf(Causes.illegal_instruction, Causes.breakpoint,
     Causes.misaligned_load, Causes.misaligned_store,
     Causes.load_access, Causes.store_access, Causes.fetch_access,
     Causes.load_page_fault, Causes.store_page_fault, Causes.fetch_page_fault)
@@ -773,7 +774,8 @@ class CSRFile(perfEventSets: EventSets = new EventSets(Seq()))(implicit p: Param
     t.insn := insn
     t.iaddr := io.pc
     t.priv := Cat(reg_debug, reg_mstatus.prv)
-    t.cause := Cat(cause(xLen-1), cause(log2Ceil(xLen)-1, 0))
+    t.cause := cause
+    t.interrupt := cause(xLen-1)
     t.tval := badaddr_value
   }
 
diff --git a/src/main/scala/rocket/DCache.scala b/src/main/scala/rocket/DCache.scala
index 48a9c5c3..7f3b3b63 100644
--- a/src/main/scala/rocket/DCache.scala
+++ b/src/main/scala/rocket/DCache.scala
@@ -204,7 +204,8 @@ class DCacheModule(outer: DCache) extends HellaCacheModule(outer) {
   val s1_all_data_ways = Vec(data.io.resp :+ dummyEncodeData(tl_out.d.bits.data))
   val s1_mask = Mux(s1_req.cmd === M_PWR, io.cpu.s1_data.mask, new StoreGen(s1_req.typ, s1_req.addr, UInt(0), wordBytes).mask)
 
-  val s2_valid = Reg(next=s1_valid_masked && !s1_sfence, init=Bool(false)) && !io.cpu.s2_xcpt.asUInt.orR
+  val s2_valid_pre_xcpt = Reg(next=s1_valid_masked && !s1_sfence, init=Bool(false))
+  val s2_valid = s2_valid_pre_xcpt && !io.cpu.s2_xcpt.asUInt.orR
   val s2_probe = Reg(next=s1_probe, init=Bool(false))
   val releaseInFlight = s1_probe || s2_probe || release_state =/= s_ready
   val s2_valid_masked = s2_valid && Reg(next = !s1_nack)
@@ -287,15 +288,19 @@ class DCacheModule(outer: DCache) extends HellaCacheModule(outer) {
   metaArb.io.in(2).bits.data.coh := Mux(s2_valid_hit, s2_new_hit_state, ClientMetadata.onReset)
   metaArb.io.in(2).bits.data.tag := s2_req.addr >> untagBits
 
-  // load reservations
+  // load reservations and TL error reporting
   val s2_lr = Bool(usingAtomics && !usingDataScratchpad) && s2_req.cmd === M_XLR
   val s2_sc = Bool(usingAtomics && !usingDataScratchpad) && s2_req.cmd === M_XSC
   val lrscCount = Reg(init=UInt(0))
+  val tl_error_valid = RegInit(false.B)
   val lrscValid = lrscCount > lrscBackoff
   val lrscAddr = Reg(UInt())
-  val s2_sc_fail = s2_sc && !(lrscValid && lrscAddr === (s2_req.addr >> blockOffBits))
-  when (s2_valid_hit && s2_lr) {
-    lrscCount := lrscCycles - 1
+  val lrscAddrMatch = lrscAddr === (s2_req.addr >> blockOffBits)
+  val s2_sc_fail = s2_sc && !(lrscValid && lrscAddrMatch)
+  val s2_tl_error = tl_error_valid && lrscAddrMatch
+  when (s2_valid_hit && s2_lr && !cached_grant_wait || s2_valid_cached_miss) {
+    tl_error_valid := false
+    lrscCount := Mux(s2_hit, lrscCycles - 1, 0.U)
     lrscAddr := s2_req.addr >> blockOffBits
   }
   when (lrscCount > 0) { lrscCount := lrscCount - 1 }
@@ -437,6 +442,7 @@ class DCacheModule(outer: DCache) extends HellaCacheModule(outer) {
       grantInProgress := true
       assert(cached_grant_wait, "A GrantData was unexpected by the dcache.")
       when(d_last) {
+        tl_error_valid := tl_out.d.bits.error
         cached_grant_wait := false
         grantInProgress := false
         blockProbeAfterGrantCount := blockProbeAfterGrantCycles - 1
@@ -488,7 +494,7 @@ class DCacheModule(outer: DCache) extends HellaCacheModule(outer) {
   // ignore backpressure from metaArb, which can only be caused by tag ECC
   // errors on hit-under-miss.  failing to write the new tag will leave the
   // line invalid, so we'll simply request the line again later.
-  metaArb.io.in(3).valid := grantIsCached && d_done
+  metaArb.io.in(3).valid := grantIsCached && d_done && !tl_out.d.bits.error
   metaArb.io.in(3).bits.write := true
   metaArb.io.in(3).bits.way_en := s2_victim_way
   metaArb.io.in(3).bits.addr := Cat(io.cpu.req.bits.addr >> untagBits, s2_req.addr(idxMSB, 0))
@@ -523,6 +529,8 @@ class DCacheModule(outer: DCache) extends HellaCacheModule(outer) {
   val s1_release_data_valid = Reg(next = dataArb.io.in(2).fire())
   val s2_release_data_valid = Reg(next = s1_release_data_valid && !releaseRejected)
   val releaseDataBeat = Cat(UInt(0), c_count) + Mux(releaseRejected, UInt(0), s1_release_data_valid + Cat(UInt(0), s2_release_data_valid))
+  val writeback_data_error = s2_data_decoded.map(_.error).reduce(_||_)
+  val writeback_data_uncorrectable = s2_data_decoded.map(_.uncorrectable).reduce(_||_)
 
   val nackResponseMessage = edge.ProbeAck(b = probe_bits, reportPermissions = TLPermissions.NtoN)
   val cleanReleaseMessage = edge.ProbeAck(b = probe_bits, reportPermissions = s2_report_param)
@@ -590,6 +598,12 @@ class DCacheModule(outer: DCache) extends HellaCacheModule(outer) {
   }
   tl_out_c.bits.address := probe_bits.address
   tl_out_c.bits.data := s2_data_corrected
+  tl_out_c.bits.error := inWriteback && {
+    val accrued = Reg(Bool())
+    val next = writeback_data_uncorrectable || (accrued && !c_first)
+    when (tl_out_c.fire()) { accrued := next }
+    next
+  }
 
   dataArb.io.in(2).valid := inWriteback && releaseDataBeat < refillCycles
   dataArb.io.in(2).bits.write := false
@@ -615,6 +629,11 @@ class DCacheModule(outer: DCache) extends HellaCacheModule(outer) {
   val s1_xcpt_valid = tlb.io.req.valid && !s1_nack
   val s1_xcpt = tlb.io.resp
   io.cpu.s2_xcpt := Mux(RegNext(s1_xcpt_valid), RegEnable(s1_xcpt, s1_valid_not_nacked), 0.U.asTypeOf(s1_xcpt))
+  when (s2_valid_pre_xcpt && s2_tl_error) {
+    assert(!s2_valid_hit && !s2_uncached)
+    when (s2_write) { io.cpu.s2_xcpt.ae.st := true }
+    when (s2_read) { io.cpu.s2_xcpt.ae.ld := true }
+  }
 
   if (usingDataScratchpad) {
     require(!usingVM) // therefore, req.phys means this is a slave-port access
@@ -718,8 +737,8 @@ class DCacheModule(outer: DCache) extends HellaCacheModule(outer) {
   {
     val (data_error, data_error_uncorrectable, data_error_addr) =
       if (usingDataScratchpad) (s2_valid_data_error, s2_data_error_uncorrectable, s2_req.addr) else {
-        (tl_out_c.valid && edge.hasData(tl_out_c.bits) && s2_data_decoded.map(_.error).reduce(_||_),
-         s2_data_decoded.map(_.uncorrectable).reduce(_||_),
+        (tl_out_c.fire() && inWriteback && writeback_data_error,
+         writeback_data_uncorrectable,
          tl_out_c.bits.address)
       }
     val error_addr =
diff --git a/src/main/scala/rocket/ICache.scala b/src/main/scala/rocket/ICache.scala
index f5abe304..9eea1958 100644
--- a/src/main/scala/rocket/ICache.scala
+++ b/src/main/scala/rocket/ICache.scala
@@ -175,15 +175,13 @@ class ICacheModule(outer: ICache) extends LazyModuleImp(outer)
   val tag_array = SeqMem(nSets, Vec(nWays, UInt(width = tECC.width(1 + tagBits))))
   val tag_rdata = tag_array.read(s0_vaddr(untagBits-1,blockOffBits), !refill_done && s0_valid)
   val accruedRefillError = Reg(Bool())
-  val refillError = tl_out.d.bits.error || (refill_cnt > 0 && accruedRefillError)
   when (refill_done) {
-    val enc_tag = tECC.encode(Cat(refillError, refill_tag))
+    val enc_tag = tECC.encode(Cat(tl_out.d.bits.error, refill_tag))
     tag_array.write(refill_idx, Vec.fill(nWays)(enc_tag), Seq.tabulate(nWays)(repl_way === _))
   }
 
   val vb_array = Reg(init=Bits(0, nSets*nWays))
   when (refill_one_beat) {
-    accruedRefillError := refillError
     // clear bit when refill starts so hit-under-miss doesn't fetch bad data
     vb_array := vb_array.bitSet(Cat(repl_way, refill_idx), refill_done && !invalidated)
   }
diff --git a/vsrc/plusarg_reader.v b/vsrc/plusarg_reader.v
index c3eb6513..0d245f9e 100644
--- a/vsrc/plusarg_reader.v
+++ b/vsrc/plusarg_reader.v
@@ -2,7 +2,7 @@
 
 // No default parameter values are intended, nor does IEEE 1800-2012 require them (clause A.2.4 param_assignment),
 // but Incisive demands them. These default values should never be used.
-module plusarg_reader #(FORMAT="borked=%d", DEFAULT=0) (
+module plusarg_reader #(parameter FORMAT="borked=%d", DEFAULT=0) (
    output [31:0] out
 );