diff --git a/litedram/common.py b/litedram/common.py
index d353954..e5b600e 100644
--- a/litedram/common.py
+++ b/litedram/common.py
@@ -56,11 +56,12 @@ def cmd_layout(aw):
     ]
 
 
-def data_layout(dw):
+def data_layout(dw, bankbits):
     return [
         ("wdata",       dw, DIR_M_TO_S),
         ("wdata_we", dw//8, DIR_M_TO_S),
-        ("rdata",       dw, DIR_S_TO_M)
+        ("rdata",       dw, DIR_S_TO_M),
+        ("rbank",   bankbits, DIR_S_TO_M)
     ]
 
 
@@ -72,7 +73,7 @@ class LiteDRAMInterface(Record):
         self.settings = settings
 
         layout = [("bank"+str(i), cmd_layout(self.aw)) for i in range(self.nbanks)]
-        layout += data_layout(self.dw)
+        layout += data_layout(self.dw, settings.geom.bankbits)
         Record.__init__(self, layout)
 
 def cmd_description(aw):
@@ -87,12 +88,12 @@ def wdata_description(dw):
         ("we",   dw//8)
     ]
 
-def rdata_description(dw):
-    return [("data", dw)]
+def rdata_description(dw, nbanks):
+    return [("data", dw), ("bank", nbanks)]
 
 
 class LiteDRAMPort:
-    def __init__(self, mode, aw, dw, cd="sys", id=0):
+    def __init__(self, mode, aw, dw, bankbits, cd="sys", id=0):
         self.mode = mode
         self.aw = aw
         self.dw = dw
@@ -103,7 +104,7 @@ class LiteDRAMPort:
 
         self.cmd = stream.Endpoint(cmd_description(aw))
         self.wdata = stream.Endpoint(wdata_description(dw))
-        self.rdata = stream.Endpoint(rdata_description(dw))
+        self.rdata = stream.Endpoint(rdata_description(dw, bankbits))
 
         self.flush = Signal()
 
diff --git a/litedram/frontend/crossbar.py b/litedram/frontend/crossbar.py
index c3f52f0..fb69230 100644
--- a/litedram/frontend/crossbar.py
+++ b/litedram/frontend/crossbar.py
@@ -33,12 +33,12 @@ class LiteDRAMCrossbar(Module):
             dw = self.dw
 
         # crossbar port
-        port = LiteDRAMPort(mode, self.rca_bits + self.bank_bits, self.dw, "sys", len(self.masters))
+        port = LiteDRAMPort(mode, self.rca_bits + self.bank_bits, self.dw, self.bank_bits, "sys", len(self.masters))
         self.masters.append(port)
 
         # clock domain crossing
         if cd != "sys":
-            new_port = LiteDRAMPort(mode, port.aw, port.dw, cd, port.id)
+            new_port = LiteDRAMPort(mode, port.aw, port.dw, self.bank_bits, cd, port.id)
             self.submodules += LiteDRAMPortCDC(new_port, port)
             port = new_port
 
@@ -48,7 +48,7 @@ class LiteDRAMCrossbar(Module):
                 adr_shift = -log2_int(dw//self.dw)
             else:
                 adr_shift = log2_int(self.dw//dw)
-            new_port = LiteDRAMPort(mode, port.aw + adr_shift, dw, cd, port.id)
+            new_port = LiteDRAMPort(mode, port.aw + adr_shift, dw, self.bank_bits, cd, port.id)
             self.submodules += ClockDomainsRenamer(cd)(LiteDRAMPortConverter(new_port, port, reverse))
             port = new_port
 
@@ -68,6 +68,8 @@ class LiteDRAMCrossbar(Module):
 
         arbiters = [roundrobin.RoundRobin(nmasters, roundrobin.SP_CE) for n in range(self.nbanks)]
         self.submodules += arbiters
+        
+        rbank = Signal(max=self.nbanks)
         for nb, arbiter in enumerate(arbiters):
             bank = getattr(controller, "bank"+str(nb))
 
@@ -89,6 +91,12 @@ class LiteDRAMCrossbar(Module):
                 arbiter.ce.eq(~bank.valid & ~bank.lock)
             ]
 
+            # Get rdata source bank
+            self.sync += \
+                If((arbiter.grant == nm) & bank.rdata_valid,
+                    rbank.eq(nb)
+                )
+
             # route requests
             self.comb += [
                 bank.adr.eq(Array(m_rca)[arbiter.grant]),
@@ -116,6 +124,12 @@ class LiteDRAMCrossbar(Module):
                     master_rdata_valid = new_master_rdata_valid
                 master_rdata_valids[nm] = master_rdata_valid
 
+        # Delay bank output to match rvalid
+        for i in range(self.read_latency-1):
+            new_master_rbank = Signal(max=self.nbanks)
+            self.sync += new_master_rbank.eq(rbank)
+            rbank = new_master_rbank
+
         for master, master_ready in zip(self.masters, master_readys):
             self.comb += master.cmd.ready.eq(master_ready)
         for master, master_wdata_ready in zip(self.masters, master_wdata_readys):
@@ -139,15 +153,18 @@ class LiteDRAMCrossbar(Module):
         # route data reads
         for master in self.masters:
             self.comb += master.rdata.data.eq(self.controller.rdata)
+            self.comb += master.rdata.bank.eq(rbank)
 
     def split_master_addresses(self, bank_bits, rca_bits, cba_shift):
         m_ba = []    # bank address
         m_rca = []    # row and column address
+        print("cba_shift: " + str(cba_shift))
         for master in self.masters:
             cba = Signal(self.bank_bits)
             rca = Signal(self.rca_bits)
             cba_upper = cba_shift + bank_bits
             self.comb += cba.eq(master.cmd.adr[cba_shift:cba_upper])
+            print("cba_shift: " + str(cba_shift) + " cba_upper: " + str(cba_upper))
             if cba_shift < self.rca_bits:
                 if cba_shift:
                     self.comb += rca.eq(Cat(master.cmd.adr[:cba_shift],
diff --git a/litedram/frontend/dma.py b/litedram/frontend/dma.py
index db24afb..1c2e1fb 100644
--- a/litedram/frontend/dma.py
+++ b/litedram/frontend/dma.py
@@ -69,7 +69,7 @@ class LiteDRAMDMAReader(Module):
         self.submodules += fifo
 
         self.comb += [
-            port.rdata.connect(fifo.sink),
+            port.rdata.connect(fifo.sink, omit=["bank"]),
             fifo.source.connect(source),
             data_dequeued.eq(source.valid & source.ready)
         ]