1 from nmigen
import Elaboratable
, Module
, Signal
, Shape
, unsigned
, Cat
, Mux
2 from nmigen
import Const
3 from soc
.fu
.mmu
.pipe_data
import MMUInputData
, MMUOutputData
, MMUPipeSpec
4 from nmutil
.singlepipe
import ControlBase
5 from nmutil
.util
import rising_edge
7 from soc
.experiment
.mmu
import MMU
8 from soc
.experiment
.dcache
import DCache
10 from soc
.decoder
.power_fields
import DecodeFields
11 from soc
.decoder
.power_fieldsn
import SignalBitRange
12 from soc
.decoder
.power_decoder2
import decode_spr_num
13 from soc
.decoder
.power_enums
import MicrOp
, SPR
, XER_bits
15 from soc
.experiment
.pimem
import PortInterface
16 from soc
.experiment
.pimem
import PortInterfaceBase
18 from soc
.experiment
.mem_types
import LoadStore1ToDCacheType
, LoadStore1ToMMUType
19 from soc
.experiment
.mem_types
import DCacheToLoadStore1Type
, MMUToLoadStore1Type
21 # for testing purposes
22 from soc
.experiment
.testmem
import TestMemory
24 # glue logic for microwatt mmu and dcache
25 class LoadStore1(PortInterfaceBase
):
26 def __init__(self
, regwid
=64, addrwid
=4):
27 super().__init
__(regwid
, addrwid
)
28 self
.d_in
= LoadStore1ToDCacheType()
29 self
.d_out
= DCacheToLoadStore1Type()
30 self
.l_in
= LoadStore1ToMMUType()
31 self
.l_out
= MMUToLoadStore1Type()
32 # for debugging with gtkwave only
33 self
.debug1
= Signal()
34 self
.debug2
= Signal()
36 def set_wr_addr(self
, m
, addr
, mask
):
37 #m.d.comb += self.d_in.valid.eq(1)
38 #m.d.comb += self.l_in.valid.eq(1)
39 #m.d.comb += self.d_in.load.eq(0)
40 #m.d.comb += self.l_in.load.eq(0)
41 m
.d
.comb
+= self
.d_in
.addr
.eq(addr
)
42 m
.d
.comb
+= self
.l_in
.addr
.eq(addr
)
46 def set_rd_addr(self
, m
, addr
, mask
):
47 m
.d
.comb
+= self
.d_in
.valid
.eq(1)
48 m
.d
.comb
+= self
.l_in
.valid
.eq(1)
49 m
.d
.comb
+= self
.d_in
.load
.eq(1)
50 m
.d
.comb
+= self
.l_in
.load
.eq(1)
51 m
.d
.comb
+= self
.d_in
.addr
.eq(addr
)
52 m
.d
.comb
+= self
.l_in
.addr
.eq(addr
)
53 m
.d
.comb
+= self
.debug1
.eq(1)
54 # m.d.comb += self.debug2.eq(1)
55 # connect testmem first
56 return None #FIXME return value
58 def set_wr_data(self
, m
, data
, wen
):
59 m
.d
.comb
+= self
.d_in
.data
.eq(data
)
64 def get_rd_data(self
, m
):
66 data
= self
.d_out
.data
69 def elaborate(self
, platform
):
70 m
= super().elaborate(platform
)
76 yield from super().ports()
79 class FSMMMUStage(ControlBase
):
80 def __init__(self
, pspec
):
85 self
.p
.data_i
= MMUInputData(pspec
)
86 self
.n
.data_o
= MMUOutputData(pspec
)
88 # incoming PortInterface
89 self
.ldst
= LoadStore1() # TODO make this depend on pspec
90 self
.pi
= self
.ldst
.pi
92 # this Function Unit is extremely unusual in that it actually stores a
93 # "thing" rather than "processes inputs and produces outputs". hence
94 # why it has to be a FSM. linking up LD/ST however is going to have
95 # to be done back in Issuer (or Core)
98 self
.dcache
= DCache()
101 # for verification of DCache
102 # XXX -- read testmem.py
103 self
.testmem
= TestMemory(regwid
, aw
, granularity
=regwid
//8, init
=False)
105 # make life a bit easier in Core
106 self
.pspec
.mmu
= self
.mmu
107 self
.pspec
.dcache
= self
.dcache
109 # debugging output for gtkw
110 self
.debug0
= Signal(4)
111 self
.debug_wb_cyc
= Signal()
112 self
.debug_wb_stb
= Signal()
113 self
.debug_wb_we
= Signal()
114 #self.debug1 = Signal(64)
115 #self.debug2 = Signal(64)
116 #self.debug3 = Signal(64)
118 # for SPR field number access
120 self
.fields
= DecodeFields(SignalBitRange
, [i
.ctx
.op
.insn
])
121 self
.fields
.create_specs()
123 def elaborate(self
, platform
):
124 m
= super().elaborate(platform
)
127 # link mmu and dcache together
128 m
.submodules
.dcache
= dcache
= self
.dcache
129 m
.submodules
.mmu
= mmu
= self
.mmu
130 m
.submodules
.ldst
= ldst
= self
.ldst
131 m
.submodules
.testmem
= testmem
= self
.testmem
132 m
.d
.comb
+= dcache
.m_in
.eq(mmu
.d_out
)
133 m
.d
.comb
+= mmu
.d_in
.eq(dcache
.m_out
)
134 l_in
, l_out
= mmu
.l_in
, mmu
.l_out
135 d_in
, d_out
= dcache
.d_in
, dcache
.d_out
136 wb_out
, wb_in
= dcache
.wb_out
, dcache
.wb_in
138 # link ldst and dcache together
139 comb
+= l_in
.eq(self
.ldst
.l_in
)
140 comb
+= self
.ldst
.l_out
.eq(l_out
)
141 comb
+= d_in
.eq(self
.ldst
.d_in
)
142 comb
+= self
.ldst
.d_out
.eq(self
.dcache
.d_out
)
145 rdport
= self
.testmem
.rdport
146 comb
+= rdport
.addr
.eq(wb_out
.adr
)
147 comb
+= wb_in
.dat
.eq(rdport
.data
)
150 wrport
= self
.testmem
.wrport
151 comb
+= wrport
.addr
.eq(wb_out
.adr
)
152 comb
+= wrport
.data
.eq(wb_out
.dat
) # write st to mem
153 comb
+= wrport
.en
.eq(wb_out
.cyc
& wb_out
.we
) # enable writes
155 # connect DCache wishbone master to debugger
156 comb
+= self
.debug_wb_cyc
.eq(wb_out
.cyc
)
157 comb
+= self
.debug_wb_stb
.eq(wb_out
.stb
)
158 comb
+= self
.debug_wb_we
.eq(wb_out
.we
)
160 comb
+= wb_in
.stall
.eq(0)
161 # testmem only takes on cycle
162 with m
.If( wb_out
.cyc
):
163 m
.d
.sync
+= wb_in
.ack
.eq( wb_out
.stb
)
165 data_i
, data_o
= self
.p
.data_i
, self
.n
.data_o
166 a_i
, b_i
, o
= data_i
.ra
, data_i
.rb
, data_o
.o
169 # TODO: link these SPRs somewhere
176 m
.d
.comb
+= self
.n
.valid_o
.eq(busy
& done
)
177 m
.d
.comb
+= self
.p
.ready_o
.eq(~busy
)
179 # take copy of X-Form SPR field
180 x_fields
= self
.fields
.FormXFX
181 spr
= Signal(len(x_fields
.SPR
))
182 comb
+= spr
.eq(decode_spr_num(x_fields
.SPR
))
184 # ok so we have to "pulse" the MMU (or dcache) rather than
185 # hold the valid hi permanently. guess what this does...
188 m
.d
.comb
+= blip
.eq(rising_edge(m
, valid
))
191 with m
.If(self
.p
.valid_i
):
192 m
.d
.sync
+= busy
.eq(1)
195 # based on the Micro-Op, we work out which of MMU or DCache
196 # should "action" the operation. one of MMU or DCache gets
197 # enabled ("valid") and we twiddle our thumbs until it
200 # FIXME: properly implement MicrOp.OP_MTSPR and MicrOp.OP_MFSPR
202 with m
.Switch(op
.insn_type
):
203 comb
+= self
.debug0
.eq(3)
204 with m
.Case(MicrOp
.OP_MTSPR
):
205 # subset SPR: first check a few bits
206 with m
.If(~spr
[9] & ~spr
[5]):
208 comb
+= dsisr
.eq(a_i
[:32])
212 # pass it over to the MMU instead
214 # blip the MMU and wait for it to complete
215 comb
+= valid
.eq(1) # start "pulse"
216 comb
+= l_in
.valid
.eq(blip
) # start
217 comb
+= l_in
.mtspr
.eq(1) # mtspr mode
218 comb
+= l_in
.sprn
.eq(spr
) # which SPR
219 comb
+= l_in
.rs
.eq(a_i
) # incoming operand (RS)
220 comb
+= done
.eq(l_out
.done
) # zzzz
222 with m
.Case(MicrOp
.OP_MFSPR
):
223 comb
+= self
.debug0
.eq(3)
224 # subset SPR: first check a few bits
225 with m
.If(~spr
[9] & ~spr
[5]):
227 comb
+= o
.data
.eq(dsisr
)
229 comb
+= o
.data
.eq(dar
)
232 # pass it over to the MMU instead
234 # blip the MMU and wait for it to complete
235 comb
+= valid
.eq(1) # start "pulse"
236 comb
+= l_in
.valid
.eq(blip
) # start
237 comb
+= l_in
.mtspr
.eq(0) # mfspr!=mtspr
238 comb
+= l_in
.sprn
.eq(spr
) # which SPR
239 comb
+= l_in
.rs
.eq(a_i
) # incoming operand (RS)
240 comb
+= o
.data
.eq(l_out
.sprval
) # SPR from MMU
241 comb
+= o
.ok
.eq(l_out
.done
) # only when l_out valid
242 comb
+= done
.eq(l_out
.done
) # zzzz
244 with m
.Case(MicrOp
.OP_DCBZ
):
245 # activate dcbz mode (spec: v3.0B p850)
246 comb
+= valid
.eq(1) # start "pulse"
247 comb
+= d_in
.valid
.eq(blip
) # start
248 comb
+= d_in
.dcbz
.eq(1) # dcbz mode
249 comb
+= d_in
.addr
.eq(a_i
+ b_i
) # addr is (RA|0) + RB
250 comb
+= done
.eq(d_out
.store_done
) # TODO
251 comb
+= self
.debug0
.eq(1)
253 with m
.Case(MicrOp
.OP_TLBIE
):
254 # pass TLBIE request to MMU (spec: v3.0B p1034)
255 # note that the spr is *not* an actual spr number, it's
256 # just that those bits happen to match with field bits
258 comb
+= valid
.eq(1) # start "pulse"
259 comb
+= l_in
.valid
.eq(blip
) # start
260 comb
+= l_in
.tlbie
.eq(1) # mtspr mode
261 comb
+= l_in
.sprn
.eq(spr
) # use sprn to send insn bits
262 comb
+= l_in
.addr
.eq(b_i
) # incoming operand (RB)
263 comb
+= done
.eq(l_out
.done
) # zzzz
264 comb
+= self
.debug0
.eq(2)
266 with m
.If(self
.n
.ready_i
& self
.n
.valid_o
):
267 m
.d
.sync
+= busy
.eq(0)