diff --git a/core.vhdl b/core.vhdl
index 6c637c7..cd2bb07 100644
--- a/core.vhdl
+++ b/core.vhdl
@@ -37,6 +37,8 @@ entity core is
         wishbone_data_in  : in wishbone_slave_out;
         wishbone_data_out : out wishbone_master_out;
 
+        wb_snoop_in     : in wishbone_master_out;
+
 	dmi_addr	: in std_ulogic_vector(3 downto 0);
 	dmi_din	        : in std_ulogic_vector(63 downto 0);
 	dmi_dout	: out std_ulogic_vector(63 downto 0);
@@ -423,6 +425,7 @@ begin
             stall_out => dcache_stall_out,
             wishbone_in => wishbone_data_in,
             wishbone_out => wishbone_data_out,
+            snoop_in => wb_snoop_in,
             log_out => log_data(170 downto 151)
             );
 
diff --git a/dcache.vhdl b/dcache.vhdl
index bb93148..9916279 100644
--- a/dcache.vhdl
+++ b/dcache.vhdl
@@ -39,6 +39,8 @@ entity dcache is
         m_in         : in MmuToDcacheType;
         m_out        : out DcacheToMmuType;
 
+        snoop_in     : in wishbone_master_out := wishbone_master_out_init;
+
 	stall_out    : out std_ulogic;
 
         wishbone_out : out wishbone_master_out;
@@ -415,6 +417,11 @@ architecture rtl of dcache is
     type tlb_plru_out_t is array(tlb_index_t) of std_ulogic_vector(TLB_WAY_BITS-1 downto 0);
     signal tlb_plru_victim : tlb_plru_out_t;
 
+    signal snoop_tag_set : cache_tags_set_t;
+    signal snoop_valid   : std_ulogic;
+    signal snoop_wrtag   : cache_tag_t;
+    signal snoop_index   : index_t;
+
     --
     -- Helper functions to decode incoming requests
     --
@@ -528,7 +535,8 @@ begin
     assert LINE_SIZE mod ROW_SIZE = 0 report "LINE_SIZE not multiple of ROW_SIZE" severity FAILURE;
     assert ispow2(LINE_SIZE)    report "LINE_SIZE not power of 2" severity FAILURE;
     assert ispow2(NUM_LINES)    report "NUM_LINES not power of 2" severity FAILURE;
-    assert ispow2(ROW_PER_LINE) report "ROW_PER_LINE not power of 2" severity FAILURE;
+    assert ispow2(ROW_PER_LINE) and ROW_PER_LINE > 1
+        report "ROW_PER_LINE not power of 2 greater than 1" severity FAILURE;
     assert (ROW_BITS = INDEX_BITS + ROW_LINEBITS)
 	report "geometry bits don't add up" severity FAILURE;
     assert (LINE_OFF_BITS = ROW_OFF_BITS + ROW_LINEBITS)
@@ -783,6 +791,24 @@ begin
         end if;
     end process;
 
+    -- Cache tag RAM second read port, for snooping
+    cache_tag_read_2 : process(clk)
+        variable addr : std_ulogic_vector(REAL_ADDR_BITS - 1 downto 0);
+    begin
+        if rising_edge(clk) then
+            addr := (others => '0');
+            addr(snoop_in.adr'left downto 0) := snoop_in.adr;
+            snoop_tag_set <= cache_tags(get_index(addr));
+            snoop_wrtag <= get_tag(addr);
+            snoop_index <= get_index(addr);
+            -- Don't snoop our own cycles
+            snoop_valid <= '0';
+            if not (r1.wb.cyc = '1' and wishbone_in.stall = '0') then
+                snoop_valid <= snoop_in.cyc and snoop_in.stb and snoop_in.we;
+            end if;
+        end if;
+    end process;
+
     -- Cache request parsing and hit detection
     dcache_request : process(all)
         variable is_hit      : std_ulogic;
@@ -1293,6 +1319,13 @@ begin
                     end if;
                 end if;
 
+                -- Do invalidations from snooped stores to memory
+                for i in way_t loop
+                    if snoop_valid = '1' and read_tag(i, snoop_tag_set) = snoop_wrtag then
+                        cache_valids(snoop_index)(i) <= '0';
+                    end if;
+                end loop;
+
                 if r1.write_tag = '1' then
                     -- Store new tag in selected way
                     for i in 0 to NUM_WAYS-1 loop
diff --git a/soc.vhdl b/soc.vhdl
index d03f114..38bd5df 100644
--- a/soc.vhdl
+++ b/soc.vhdl
@@ -133,6 +133,7 @@ architecture behaviour of soc is
     -- Wishbone master (output of arbiter):
     signal wb_master_in       : wishbone_slave_out;
     signal wb_master_out      : wishbone_master_out;
+    signal wb_snoop           : wishbone_master_out;
 
     -- Main "IO" bus, from main slave decoder to the latch
     signal wb_io_in     : wishbone_master_out;
@@ -284,6 +285,7 @@ begin
 	    wishbone_insn_out => wishbone_icore_out,
 	    wishbone_data_in => wishbone_dcore_in,
 	    wishbone_data_out => wishbone_dcore_out,
+            wb_snoop_in => wb_snoop,
 	    dmi_addr => dmi_addr(3 downto 0),
 	    dmi_dout => dmi_core_dout,
 	    dmi_din => dmi_dout,
@@ -313,6 +315,18 @@ begin
 	    wb_slave_in => wb_master_in
 	    );
 
+    -- Snoop bus going to caches.
+    -- Gate stb with stall so the caches don't see the stalled strobes.
+    -- That way if the caches see a strobe when their wishbone is stalled,
+    -- they know it is an access by another master.
+    process(all)
+    begin
+        wb_snoop <= wb_master_out;
+        if wb_master_in.stall = '1' then
+            wb_snoop.stb <= '0';
+        end if;
+    end process;
+
     -- Top level Wishbone slaves address decoder & mux
     --
     -- From CPU to BRAM, DRAM, IO, selected on top 3 bits and dram_at_0