X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fsoc%2Fsimple%2Fissuer.py;h=27afb68ed4546c2751f87f35db7373f845c7464b;hb=50adde0faa929c2c1e9f305357cbae6d9f0deae7;hp=485d7e0007d0750967bbcfe015268c6d32861f77;hpb=8d375b8db0fa9ba7ad2898a35bc0a12b1e69857e;p=soc.git

diff --git a/src/soc/simple/issuer.py b/src/soc/simple/issuer.py
index 485d7e00..27afb68e 100644
--- a/src/soc/simple/issuer.py
+++ b/src/soc/simple/issuer.py
@@ -23,31 +23,70 @@ import sys
 
 from soc.decoder.power_decoder import create_pdecode
 from soc.decoder.power_decoder2 import PowerDecode2
+from soc.decoder.decode2execute1 import IssuerDecode2ToOperand
 from soc.decoder.decode2execute1 import Data
 from soc.experiment.testmem import TestMemory # test only for instructions
-from soc.regfile.regfiles import StateRegs
+from soc.regfile.regfiles import StateRegs, FastRegs
 from soc.simple.core import NonProductionCore
 from soc.config.test.test_loadstore import TestMemPspec
 from soc.config.ifetch import ConfigFetchUnit
 from soc.decoder.power_enums import MicrOp
 from soc.debug.dmi import CoreDebug, DMIInterface
+from soc.debug.jtag import JTAG
+from soc.config.pinouts import get_pinspecs
 from soc.config.state import CoreState
+from soc.interrupts.xics import XICS_ICP, XICS_ICS
+from soc.bus.simple_gpio import SimpleGPIO
+from soc.clock.select import ClockSelect, DummyPLL
+
 
 from nmutil.util import rising_edge
 
 
-class TestIssuer(Elaboratable):
+class TestIssuerInternal(Elaboratable):
     """TestIssuer - reads instructions from TestMemory and issues them
 
     efficiency and speed is not the main goal here: functional correctness is.
     """
     def __init__(self, pspec):
-        # main instruction core
+
+        # JTAG interface.  add this right at the start because if it's
+        # added it *modifies* the pspec, by adding enable/disable signals
+        # for parts of the rest of the core
+        self.jtag_en = hasattr(pspec, "debug") and pspec.debug == 'jtag'
+        if self.jtag_en:
+            subset = {'uart', 'mtwi', 'eint', 'gpio', 'mspi0', 'mspi1',
+                      'pwm', 'sd0', 'sdr'}
+            self.jtag = JTAG(get_pinspecs(subset=subset))
+            # add signals to pspec to enable/disable icache and dcache
+            # (or data and intstruction wishbone if icache/dcache not included)
+            # https://bugs.libre-soc.org/show_bug.cgi?id=520
+            # TODO: do we actually care if these are not domain-synchronised?
+            # honestly probably not.
+            pspec.wb_icache_en = self.jtag.wb_icache_en
+            pspec.wb_dcache_en = self.jtag.wb_dcache_en
+
+        # add interrupt controller?
+        self.xics = hasattr(pspec, "xics") and pspec.xics == True
+        if self.xics:
+            self.xics_icp = XICS_ICP()
+            self.xics_ics = XICS_ICS()
+            self.int_level_i = self.xics_ics.int_level_i
+
+        # add GPIO peripheral?
+        self.gpio = hasattr(pspec, "gpio") and pspec.gpio == True
+        if self.gpio:
+            self.simple_gpio = SimpleGPIO()
+            self.gpio_o = self.simple_gpio.gpio_o
+
+        # main instruction core25
         self.core = core = NonProductionCore(pspec)
 
-        # instruction decoder
+        # instruction decoder.  goes into Trap Record
         pdecode = create_pdecode()
-        self. pdecode2 = PowerDecode2(pdecode)   # decoder
+        self.cur_state = CoreState("cur") # current state (MSR/PC/EINT)
+        self.pdecode2 = PowerDecode2(pdecode, state=self.cur_state,
+                                     opkls=IssuerDecode2ToOperand)
 
         # Test Instruction memory
         self.imem = ConfigFetchUnit(pspec).fu
@@ -65,14 +104,19 @@ class TestIssuer(Elaboratable):
         self.busy_o = Signal(reset_less=True)
         self.memerr_o = Signal(reset_less=True)
 
-        # FAST regfile read /write ports for PC and MSR
-        self.state_r_pc = self.core.regs.rf['state'].r_ports['cia'] # PC rd
-        self.state_w_pc = self.core.regs.rf['state'].w_ports['d_wr1'] # PC wr
-        self.state_r_msr = self.core.regs.rf['state'].r_ports['msr'] # MSR rd
+        # FAST regfile read /write ports for PC, MSR, DEC/TB
+        staterf = self.core.regs.rf['state']
+        self.state_r_pc = staterf.r_ports['cia'] # PC rd
+        self.state_w_pc = staterf.w_ports['d_wr1'] # PC wr
+        self.state_r_msr = staterf.r_ports['msr'] # MSR rd
 
         # DMI interface access
         intrf = self.core.regs.rf['int']
+        crrf = self.core.regs.rf['cr']
+        xerrf = self.core.regs.rf['xer']
         self.int_r = intrf.r_ports['dmi'] # INT read
+        self.cr_r = crrf.r_ports['full_cr_dbg'] # CR read
+        self.xer_r = xerrf.r_ports['full_xer'] # XER read
 
         # hack method of keeping an eye on whether branch/trap set the PC
         self.state_nia = self.core.regs.rf['state'].w_ports['nia']
@@ -85,14 +129,37 @@ class TestIssuer(Elaboratable):
         m.submodules.core = core = DomainRenamer("coresync")(self.core)
         m.submodules.imem = imem = self.imem
         m.submodules.dbg = dbg = self.dbg
+        if self.jtag_en:
+            m.submodules.jtag = jtag = self.jtag
+            # TODO: UART2GDB mux, here, from external pin
+            # see https://bugs.libre-soc.org/show_bug.cgi?id=499
+            sync += dbg.dmi.connect_to(jtag.dmi)
+
+        cur_state = self.cur_state
+
+        # XICS interrupt handler
+        if self.xics:
+            m.submodules.xics_icp = icp = self.xics_icp
+            m.submodules.xics_ics = ics = self.xics_ics
+            comb += icp.ics_i.eq(ics.icp_o)           # connect ICS to ICP
+            sync += cur_state.eint.eq(icp.core_irq_o) # connect ICP to core
+
+        # GPIO test peripheral
+        if self.gpio:
+            m.submodules.simple_gpio = simple_gpio = self.simple_gpio
+
+        # connect one GPIO output to ICS bit 15 (like in microwatt soc.vhdl)
+        # XXX causes litex ECP5 test to get wrong idea about input and output
+        # (but works with verilator sim *sigh*)
+        #if self.gpio and self.xics:
+        #   comb += self.int_level_i[15].eq(simple_gpio.gpio_o[0])
 
         # instruction decoder
         pdecode = create_pdecode()
         m.submodules.dec2 = pdecode2 = self.pdecode2
 
         # convenience
-        dmi = dbg.dmi
-        d_reg = dbg.dbg_gpr
+        dmi, d_reg, d_cr, d_xer, = dbg.dmi, dbg.d_gpr, dbg.d_cr, dbg.d_xer
         intrf = self.core.regs.rf['int']
 
         # clock delay power-on reset
@@ -101,21 +168,21 @@ class TestIssuer(Elaboratable):
         core_sync = ClockDomain("coresync")
         m.domains += cd_por, cd_sync, core_sync
 
-        delay = Signal(range(4), reset=1)
+        ti_rst = Signal(reset_less=True)
+        delay = Signal(range(4), reset=3)
         with m.If(delay != 0):
             m.d.por += delay.eq(delay - 1)
         comb += cd_por.clk.eq(ClockSignal())
-        comb += core_sync.clk.eq(ClockSignal())
-        # XXX TODO: power-on reset delay (later)
-        #comb += core.core_reset_i.eq(delay != 0 | dbg.core_rst_o)
+
+        # power-on reset delay 
+        core_rst = ResetSignal("coresync")
+        comb += ti_rst.eq(delay != 0 | dbg.core_rst_o | ResetSignal())
+        comb += core_rst.eq(ti_rst)
 
         # busy/halted signals from core
         comb += self.busy_o.eq(core.busy_o)
         comb += pdecode2.dec.bigendian.eq(self.core_bigendian_i)
 
-        # current state (MSR/PC at the moment
-        cur_state = CoreState("cur")
-
         # temporary hack: says "go" immediately for both address gen and ST
         l0 = core.l0
         ldst = core.fus.fus['ldst0']
@@ -134,41 +201,40 @@ class TestIssuer(Elaboratable):
 
         # read the PC
         pc = Signal(64, reset_less=True)
+        pc_ok_delay = Signal()
+        sync += pc_ok_delay.eq(~self.pc_i.ok)
         with m.If(self.pc_i.ok):
             # incoming override (start from pc_i)
             comb += pc.eq(self.pc_i.data)
         with m.Else():
-            # otherwise read StateRegs regfile for PC
+            # otherwise read StateRegs regfile for PC...
             comb += self.state_r_pc.ren.eq(1<<StateRegs.PC)
+        # ... but on a 1-clock delay
+        with m.If(pc_ok_delay):
             comb += pc.eq(self.state_r_pc.data_o)
 
         # don't write pc every cycle
-        sync += self.state_w_pc.wen.eq(0)
-        sync += self.state_w_pc.data_i.eq(0)
+        comb += self.state_w_pc.wen.eq(0)
+        comb += self.state_w_pc.data_i.eq(0)
 
         # don't read msr every cycle
-        sync += self.state_r_msr.ren.eq(0)
+        comb += self.state_r_msr.ren.eq(0)
+        msr_read = Signal(reset=1)
 
         # connect up debug signals
         # TODO comb += core.icache_rst_i.eq(dbg.icache_rst_o)
-        comb += core.core_stopped_i.eq(dbg.core_stop_o)
-        comb += core.core_reset_i.eq(dbg.core_rst_o)
         comb += dbg.terminate_i.eq(core.core_terminate_o)
         comb += dbg.state.pc.eq(pc)
+        #comb += dbg.state.pc.eq(cur_state.pc)
         comb += dbg.state.msr.eq(cur_state.msr)
 
-        # temporarily connect up core execute decode to pdecode2
-        comb += core.e.eq(pdecode2.e)
-
         # temporaries
-        core_busy_o = core.busy_o                  # core is busy
-        core_ivalid_i = core.ivalid_i              # instruction is valid
-        core_issue_i = core.issue_i                # instruction is issued
-        core_be_i = pdecode2.dec.bigendian         # bigendian mode
-        core_opcode_i = pdecode2.dec.raw_opcode_in # raw opcode
+        core_busy_o = core.busy_o                 # core is busy
+        core_ivalid_i = core.ivalid_i             # instruction is valid
+        core_issue_i = core.issue_i               # instruction is issued
+        dec_opcode_i = pdecode2.dec.raw_opcode_in # raw opcode
 
-        insn_type = pdecode2.e.do.insn_type
-        insn_state = pdecode2.state
+        insn_type = core.e.do.insn_type
 
         # actually use a nmigen FSM for the first time (w00t)
         # this FSM is perhaps unusual in that it detects conditions
@@ -180,7 +246,10 @@ class TestIssuer(Elaboratable):
             # waiting (zzz)
             with m.State("IDLE"):
                 sync += pc_changed.eq(0)
-                with m.If(~dbg.core_stop_o):
+                sync += core.e.eq(0)
+                sync += core.raw_insn_i.eq(0)
+                sync += core.bigendian_i.eq(0)
+                with m.If(~dbg.core_stop_o & ~core_rst):
                     # instruction allowed to go: start by reading the PC
                     # capture the PC and also drop it into Insn Memory
                     # we have joined a pair of combinatorial memory
@@ -190,17 +259,22 @@ class TestIssuer(Elaboratable):
                     comb += self.imem.f_valid_i.eq(1)
                     sync += cur_state.pc.eq(pc)
 
-                    # read MSR, latch it, and put it in decode "state"
-                    sync += self.state_r_msr.ren.eq(1<<StateRegs.MSR)
-                    sync += cur_state.msr.eq(self.state_r_msr.data_o)
+                    # initiate read of MSR.  arrives one clock later
+                    comb += self.state_r_msr.ren.eq(1<<StateRegs.MSR)
+                    sync += msr_read.eq(0)
 
                     m.next = "INSN_READ" # move to "wait for bus" phase
+                with m.Else():
+                    comb += core.core_stopped_i.eq(1)
+                    comb += dbg.core_stopped_i.eq(1)
 
             # dummy pause to find out why simulation is not keeping up
             with m.State("INSN_READ"):
-                with m.If(dbg.core_stop_o):
-                    m.next = "IDLE" # back to idle
-                with m.Elif(self.imem.f_busy_o): # zzz...
+                # one cycle later, msr read arrives.  valid only once.
+                with m.If(~msr_read):
+                    sync += msr_read.eq(1) # yeah don't read it again
+                    sync += cur_state.msr.eq(self.state_r_msr.data_o)
+                with m.If(self.imem.f_busy_o): # zzz...
                     # busy: stay in wait-read
                     comb += self.imem.a_valid_i.eq(1)
                     comb += self.imem.f_valid_i.eq(1)
@@ -211,18 +285,19 @@ class TestIssuer(Elaboratable):
                         insn = f_instr_o
                     else:
                         insn = f_instr_o.word_select(cur_state.pc[2], 32)
-                    comb += core_opcode_i.eq(insn) # actual opcode
+                    comb += dec_opcode_i.eq(insn) # actual opcode
+                    sync += core.e.eq(pdecode2.e)
+                    sync += core.state.eq(cur_state)
+                    sync += core.raw_insn_i.eq(dec_opcode_i)
+                    sync += core.bigendian_i.eq(self.core_bigendian_i)
                     sync += ilatch.eq(insn) # latch current insn
+                    # also drop PC and MSR into decode "state"
                     m.next = "INSN_START" # move to "start"
 
             # waiting for instruction bus (stays there until not busy)
             with m.State("INSN_START"):
                 comb += core_ivalid_i.eq(1) # instruction is valid
                 comb += core_issue_i.eq(1)  # and issued
-                comb += core_opcode_i.eq(ilatch) # actual opcode
-
-                # also drop PC and MSR into decode "state"
-                comb += insn_state.eq(cur_state)
 
                 m.next = "INSN_ACTIVE" # move to "wait completion"
 
@@ -230,22 +305,22 @@ class TestIssuer(Elaboratable):
             with m.State("INSN_ACTIVE"):
                 with m.If(insn_type != MicrOp.OP_NOP):
                     comb += core_ivalid_i.eq(1) # instruction is valid
-                comb += core_opcode_i.eq(ilatch) # actual opcode
-                comb += insn_state.eq(cur_state)     # and MSR and PC
-                with m.If(self.state_nia.wen):
+                with m.If(self.state_nia.wen & (1<<StateRegs.PC)):
                     sync += pc_changed.eq(1)
                 with m.If(~core_busy_o): # instruction done!
                     # ok here we are not reading the branch unit.  TODO
                     # this just blithely overwrites whatever pipeline
                     # updated the PC
                     with m.If(~pc_changed):
-                        sync += self.state_w_pc.wen.eq(1<<StateRegs.PC)
-                        sync += self.state_w_pc.data_i.eq(nia)
+                        comb += self.state_w_pc.wen.eq(1<<StateRegs.PC)
+                        comb += self.state_w_pc.data_i.eq(nia)
+                    sync += core.e.eq(0)
+                    sync += core.raw_insn_i.eq(0)
+                    sync += core.bigendian_i.eq(0)
                     m.next = "IDLE" # back to idle
 
         # this bit doesn't have to be in the FSM: connect up to read
         # regfiles on demand from DMI
-
         with m.If(d_reg.req): # request for regfile access being made
             # TODO: error-check this
             # XXX should this be combinatorial?  sync better?
@@ -254,9 +329,88 @@ class TestIssuer(Elaboratable):
             else:
                 comb += self.int_r.addr.eq(d_reg.addr)
                 comb += self.int_r.ren.eq(1)
+        d_reg_delay  = Signal()
+        sync += d_reg_delay.eq(d_reg.req)
+        with m.If(d_reg_delay):
+            # data arrives one clock later
             comb += d_reg.data.eq(self.int_r.data_o)
             comb += d_reg.ack.eq(1)
 
+        # sigh same thing for CR debug
+        with m.If(d_cr.req): # request for regfile access being made
+            comb += self.cr_r.ren.eq(0b11111111) # enable all
+        d_cr_delay  = Signal()
+        sync += d_cr_delay.eq(d_cr.req)
+        with m.If(d_cr_delay):
+            # data arrives one clock later
+            comb += d_cr.data.eq(self.cr_r.data_o)
+            comb += d_cr.ack.eq(1)
+
+        # aaand XER...
+        with m.If(d_xer.req): # request for regfile access being made
+            comb += self.xer_r.ren.eq(0b111111) # enable all
+        d_xer_delay  = Signal()
+        sync += d_xer_delay.eq(d_xer.req)
+        with m.If(d_xer_delay):
+            # data arrives one clock later
+            comb += d_xer.data.eq(self.xer_r.data_o)
+            comb += d_xer.ack.eq(1)
+
+        # DEC and TB inc/dec FSM
+        self.tb_dec_fsm(m, cur_state.dec)
+
+        return m
+
+    def tb_dec_fsm(self, m, spr_dec):
+        """tb_dec_fsm
+
+        this is a FSM for updating either dec or tb.  it runs alternately
+        DEC, TB, DEC, TB.  note that SPR pipeline could have written a new
+        value to DEC, however the regfile has "passthrough" on it so this
+        *should* be ok.
+
+        see v3.0B p1097-1099 for Timeer Resource and p1065 and p1076
+        """
+
+        comb, sync = m.d.comb, m.d.sync
+        fast_rf = self.core.regs.rf['fast']
+        fast_r_dectb = fast_rf.r_ports['issue'] # DEC/TB
+        fast_w_dectb = fast_rf.w_ports['issue'] # DEC/TB
+
+        with m.FSM() as fsm:
+
+            # initiates read of current DEC
+            with m.State("DEC_READ"):
+                comb += fast_r_dectb.addr.eq(FastRegs.DEC)
+                comb += fast_r_dectb.ren.eq(1)
+                m.next = "DEC_WRITE"
+
+            # waits for DEC read to arrive (1 cycle), updates with new value
+            with m.State("DEC_WRITE"):
+                new_dec = Signal(64)
+                # TODO: MSR.LPCR 32-bit decrement mode
+                comb += new_dec.eq(fast_r_dectb.data_o - 1)
+                comb += fast_w_dectb.addr.eq(FastRegs.DEC)
+                comb += fast_w_dectb.wen.eq(1)
+                comb += fast_w_dectb.data_i.eq(new_dec)
+                sync += spr_dec.eq(new_dec) # copy into cur_state for decoder
+                m.next = "TB_READ"
+
+            # initiates read of current TB
+            with m.State("TB_READ"):
+                comb += fast_r_dectb.addr.eq(FastRegs.TB)
+                comb += fast_r_dectb.ren.eq(1)
+                m.next = "TB_WRITE"
+
+            # waits for read TB to arrive, initiates write of current TB
+            with m.State("TB_WRITE"):
+                new_tb = Signal(64)
+                comb += new_tb.eq(fast_r_dectb.data_o + 1)
+                comb += fast_w_dectb.addr.eq(FastRegs.TB)
+                comb += fast_w_dectb.wen.eq(1)
+                comb += fast_w_dectb.data_i.eq(new_tb)
+                m.next = "DEC_READ"
+
         return m
 
     def __iter__(self):
@@ -272,21 +426,99 @@ class TestIssuer(Elaboratable):
         return list(self)
 
     def external_ports(self):
-        return self.pc_i.ports() + [self.pc_o,
-                                    self.memerr_o,
-                                    self.core_bigendian_i,
-                                    ClockSignal(),
-                                    ResetSignal(),
-                                    self.busy_o,
-                                    ] + \
-                list(self.dbg.dmi.ports()) + \
-                list(self.imem.ibus.fields.values()) + \
-                list(self.core.l0.cmpi.lsmem.lsi.dbus.fields.values())
+        ports = self.pc_i.ports()
+        ports += [self.pc_o, self.memerr_o, self.core_bigendian_i, self.busy_o,
+                ]
+
+        if self.jtag_en:
+            ports += list(self.jtag.external_ports())
+        else:
+            # don't add DMI if JTAG is enabled
+            ports += list(self.dbg.dmi.ports())
+
+        ports += list(self.imem.ibus.fields.values())
+        ports += list(self.core.l0.cmpi.lsmem.lsi.slavebus.fields.values())
+
+        if self.xics:
+            ports += list(self.xics_icp.bus.fields.values())
+            ports += list(self.xics_ics.bus.fields.values())
+            ports.append(self.int_level_i)
+
+        if self.gpio:
+            ports += list(self.simple_gpio.bus.fields.values())
+            ports.append(self.gpio_o)
+
+        return ports
 
     def ports(self):
         return list(self)
 
 
+class TestIssuer(Elaboratable):
+    def __init__(self, pspec):
+        self.ti = TestIssuerInternal(pspec)
+
+        self.pll = DummyPLL()
+        self.clksel = ClockSelect()
+
+        # PLL direct clock or not
+        self.pll_en = hasattr(pspec, "use_pll") and pspec.use_pll
+
+    def elaborate(self, platform):
+        m = Module()
+        comb = m.d.comb
+
+        # TestIssuer runs at direct clock
+        m.submodules.ti = ti = self.ti
+        cd_int = ClockDomain("coresync")
+
+        # ClockSelect runs at PLL output internal clock rate
+        m.submodules.clksel = clksel = DomainRenamer("pllclk")(self.clksel)
+        m.submodules.pll = pll = self.pll
+
+        # add 2 clock domains established above...
+        cd_pll = ClockDomain("pllclk")
+        m.domains += cd_pll
+
+        # internal clock is set to selector clock-out.  has the side-effect of
+        # running TestIssuer at this speed (see DomainRenamer("intclk") above)
+        intclk = ClockSignal("coresync")
+        if self.pll_en:
+            comb += intclk.eq(clksel.core_clk_o)
+        else:
+            comb += intclk.eq(ClockSignal())
+
+        # PLL clock established.  has the side-effect of running clklsel
+        # at the PLL's speed (see DomainRenamer("pllclk") above)
+        pllclk = ClockSignal("pllclk")
+        comb += pllclk.eq(pll.clk_pll_o)
+
+        # wire up external 24mhz to PLL and clksel
+        comb += clksel.clk_24_i.eq(ClockSignal())
+        comb += pll.clk_24_i.eq(clksel.clk_24_i)
+
+        # now wire up ResetSignals.  don't mind them all being in this domain
+        #int_rst = ResetSignal("coresync")
+        pll_rst = ResetSignal("pllclk")
+        #comb += int_rst.eq(ResetSignal())
+        comb += pll_rst.eq(ResetSignal())
+
+        return m
+
+    def ports(self):
+        return list(self.ti.ports()) + list(self.pll.ports()) + \
+               [ClockSignal(), ResetSignal()] + \
+               list(self.clksel.ports())
+
+    def external_ports(self):
+        ports = self.ti.external_ports()
+        ports.append(ClockSignal())
+        ports.append(ResetSignal())
+        ports.append(self.clksel.clk_sel_i)
+        ports.append(self.clksel.pll_48_o)
+        return ports
+
+
 if __name__ == '__main__':
     units = {'alu': 1, 'cr': 1, 'branch': 1, 'trap': 1, 'logical': 1,
              'spr': 1,