mergeA: fix 2 bugs about merge permission_check and pftrain vaddr

src/main/scala/coupledL2/Common.scala

@@ -42,6 +42,7 @@ trait HasChannelBits { this: Bundle =>
 class MergeTaskBundle(implicit p: Parameters) extends L2Bundle {
   val off = UInt(offsetBits.W)
   val alias = aliasBitsOpt.map(_ => UInt(aliasBitsOpt.get.W)) // color bits in cache-alias issue
+  val vaddr = vaddrBitsOpt.map(_ => UInt(vaddrBitsOpt.get.W)) // vaddr passed by client cache, for prefetcher train
   val opcode = UInt(3.W) // type of the task operation
   val param = UInt(3.W)
   val sourceId = UInt(sourceIdBits.W) // tilelink sourceID
@@ -179,6 +180,7 @@ class MSHRInfo(implicit p: Parameters) extends L2Bundle {
 
   // whether the mshr_task already in mainpipe
   val s_refill = Bool()
+  val param = UInt(3.W)
   val mergeA = Bool() // whether the mshr already merge an acquire(avoid alias merge)
 }
 

src/main/scala/coupledL2/MSHR.scala

@@ -399,6 +399,7 @@ class MSHR(implicit p: Parameters) extends L2Module {
     val merge_task = Mux(io.aMergeTask.valid, io.aMergeTask.bits, merge_task_r)
     mp_grant.aMergeTask.off := merge_task.off
     mp_grant.aMergeTask.alias.foreach(_ := merge_task.alias.getOrElse(0.U))
+    mp_grant.aMergeTask.vaddr.foreach(_ := merge_task.vaddr.getOrElse(0.U))
     mp_grant.aMergeTask.opcode := odOpGen(merge_task.opcode)
     mp_grant.aMergeTask.param := MuxLookup( // Acquire -> Grant
       merge_task.param,
@@ -585,6 +586,7 @@ class MSHR(implicit p: Parameters) extends L2Module {
   io.msInfo.bits.isAcqOrPrefetch := req_acquire || req_prefetch
   io.msInfo.bits.isPrefetch := req_prefetch
   io.msInfo.bits.s_refill := state.s_refill
+  io.msInfo.bits.param := req.param
   io.msInfo.bits.mergeA := mergeA
 
   assert(!(c_resp.valid && !io.status.bits.w_c_resp))

src/main/scala/coupledL2/MainPipe.scala

@@ -416,7 +416,7 @@ class MainPipe(implicit p: Parameters) extends L2Module {
       train.bits.set := req_s3.set
       train.bits.needT := Mux(req_s3.mergeA, needT(req_s3.aMergeTask.opcode, req_s3.aMergeTask.param),req_needT_s3)
       train.bits.source := Mux(req_s3.mergeA, req_s3.aMergeTask.sourceId, req_s3.sourceId)
-      train.bits.vaddr.foreach(_ := req_s3.vaddr.getOrElse(0.U))
+      train.bits.vaddr.foreach(_ := Mux(req_s3.mergeA, req_s3.aMergeTask.vaddr.getOrElse(0.U), req_s3.vaddr.getOrElse(0.U)))
   }
 
   /* ======== Stage 4 ======== */

src/main/scala/coupledL2/RequestBuffer.scala

@@ -2,6 +2,7 @@ package coupledL2
 
 import chipsalliance.rocketchip.config.Parameters
 import freechips.rocketchip.tilelink.TLMessages._
+import freechips.rocketchip.tilelink.TLPermissions._
 import chisel3._
 import chisel3.util._
 import coupledL2.utils._
@@ -120,7 +121,7 @@ class RequestBuffer(flow: Boolean = true, entries: Int = 4)(implicit p: Paramete
   // incoming Acquire can be merged with late_pf MSHR block
   val mergeAMask = VecInit(io.mshrInfo.map(s =>
     s.valid && s.bits.isPrefetch && sameAddr(in, s.bits) && !s.bits.willFree && !s.bits.dirHit && !s.bits.s_refill &&
-      in.fromA && (in.opcode === AcquireBlock || in.opcode === AcquirePerm) && !s.bits.mergeA
+      in.fromA && (in.opcode === AcquireBlock || in.opcode === AcquirePerm) && !s.bits.mergeA && !(in.param === toT && s.bits.param === toB)
   )).asUInt
   val mergeA = mergeAMask.orR
   val mergeAId = OHToUInt(mergeAMask)