gateware.spi: refactor, sim verified

This commit is contained in:
Robert Jördens 2016-02-28 20:40:06 +01:00
parent bd9ceb4e12
commit 3b6999ac06
1 changed files with 283 additions and 171 deletions

View File

@ -1,11 +1,158 @@
from itertools import product from itertools import product
from migen import * from migen import *
from migen.genlib.fsm import * from migen.genlib.fsm import FSM, NextState
from migen.genlib.misc import WaitTimer
from misoc.interconnect import wishbone from misoc.interconnect import wishbone
class SPIClockGen(Module):
def __init__(self, width):
self.load = Signal(width)
self.bias = Signal() # bias this clock phase to longer times
self.edge = Signal()
self.clk = Signal(reset=1)
cnt = Signal.like(self.load)
self.comb += [
self.edge.eq(cnt == 0),
]
self.sync += [
cnt.eq(cnt - 1),
If(self.edge,
cnt.eq(self.load[1:] +
(self.load[0] & self.bias)),
self.clk.eq(~self.clk),
)
]
class SPIRegister(Module):
def __init__(self, width):
self.data = Signal(width)
self.o = Signal()
self.i = Signal()
self.lsb = Signal()
self.shift = Signal()
self.sample = Signal()
self.comb += [
self.o.eq(Mux(self.lsb, self.data[0], self.data[-1])),
]
self.sync += [
If(self.shift,
If(self.lsb,
self.data[:-1].eq(self.data[1:]),
).Else(
self.data[1:].eq(self.data[:-1]),
)
),
If(self.sample,
If(self.lsb,
self.data[-1].eq(self.i),
).Else(
self.data[0].eq(self.i),
)
)
]
class SPIBitCounter(Module):
def __init__(self, width):
self.n_read = Signal(width)
self.n_write = Signal(width)
self.read = Signal()
self.write = Signal()
self.done = Signal()
self.comb += [
self.write.eq(self.n_write != 0),
self.read.eq(self.n_read != 0),
self.done.eq(~(self.write | self.read)),
]
self.sync += [
If(self.write,
self.n_write.eq(self.n_write - 1),
).Elif(self.read,
self.n_read.eq(self.n_read - 1),
)
]
class SPIMachine(Module):
def __init__(self, data_width, clock_width, bits_width):
ce = CEInserter()
self.submodules.cg = ce(SPIClockGen(clock_width))
self.submodules.reg = ce(SPIRegister(data_width))
self.submodules.bits = ce(SPIBitCounter(bits_width))
self.div_write = Signal.like(self.cg.load)
self.div_read = Signal.like(self.cg.load)
self.clk_phase = Signal()
self.start = Signal()
self.cs = Signal()
self.oe = Signal()
self.done = Signal()
# # #
fsm = CEInserter()(FSM("IDLE"))
self.submodules += fsm
fsm.act("IDLE",
If(self.start,
If(self.clk_phase,
NextState("WAIT"),
).Else(
NextState("SETUP"),
)
)
)
fsm.act("SETUP",
self.reg.sample.eq(1),
NextState("HOLD"),
)
fsm.act("HOLD",
If(self.bits.done & ~self.start,
If(self.clk_phase,
NextState("IDLE"),
).Else(
NextState("WAIT"),
)
).Else(
self.reg.shift.eq(1),
NextState("SETUP"),
)
)
fsm.act("WAIT",
If(self.bits.done,
NextState("IDLE"),
).Else(
NextState("SETUP"),
)
)
write0 = Signal()
self.sync += [
If(self.cg.edge & self.reg.shift,
write0.eq(self.bits.write),
)
]
self.comb += [
self.cg.ce.eq(self.start | self.cs | ~self.cg.edge),
If(self.bits.write | ~self.bits.read,
self.cg.load.eq(self.div_write),
).Else(
self.cg.load.eq(self.div_read),
),
self.cg.bias.eq(fsm.before_entering("SETUP")),
fsm.ce.eq(self.cg.edge),
self.cs.eq(~fsm.ongoing("IDLE")),
self.reg.ce.eq(self.cg.edge),
self.bits.ce.eq(self.cg.edge & self.reg.sample),
self.done.eq(self.cg.edge & self.bits.done & fsm.ongoing("HOLD")),
self.oe.eq(write0 | self.bits.write),
]
class SPIMaster(Module): class SPIMaster(Module):
"""SPI Master. """SPI Master.
@ -15,50 +162,61 @@ class SPIMaster(Module):
* If there is a miso wire in pads, the input and output can be done * If there is a miso wire in pads, the input and output can be done
with two signals (a.k.a. 4-wire SPI), else mosi must be used for with two signals (a.k.a. 4-wire SPI), else mosi must be used for
both output and input (a.k.a. 3-wire SPI) and config.half_duplex both output and input (a.k.a. 3-wire SPI) and config.half_duplex
needs to be set. must to be set when reading data is desired.
* Every transfer consists of a 0-M bit write followed by a 0-M * Every transfer consists of a write_length 0-M bit write followed
bit read. by a read_length 0-M bit read.
* cs_n is always asserted at the beginning and deasserted * cs_n is asserted at the beginning and deasserted at the end of the
at the end of the transfer. transfer if there is no other transfer pending.
* cs_n handling is agnostic to whether it is one-hot or decoded * cs_n handling is agnostic to whether it is one-hot or decoded
somewhere downstream. If it is decoded, "cs_n all deasserted" somewhere downstream. If it is decoded, "cs_n all deasserted"
should be handled accordingly (no slave selected). should be handled accordingly (no slave selected).
If it is one-hot, asserting multiple slaves should only be attempted If it is one-hot, asserting multiple slaves should only be attempted
if miso is either not connected between slaves or open collector. if miso is either not connected between slaves, or open collector,
cs can also be handled independently through other means. or correctly multiplexed externally.
* If config.cs_polarity == 0 (cs active low, the default), * If config.cs_polarity == 0 (cs active low, the default),
"cs_n all deasserted" means "all cs_n bits high". "cs_n all deasserted" means "all cs_n bits high".
* cs is not mandatory in pads. Framing and chip selection can also
be handled independently through other means.
* The first bit output on mosi is always the MSB/LSB (depending on * The first bit output on mosi is always the MSB/LSB (depending on
config.lsb_first) of the data register, independent of config.lsb_first) of the data register, independent of
xfer.write_len. The last bit input from miso always ends up in xfer.write_len. The last bit input from miso always ends up in
the LSB/MSB (respectively) of the data register, independent of the LSB/MSB (respectively) of the data register, independent of
read_len. read_len.
* For 4-wire SPI only the sum of read_len and write_len matters. The * For 4-wire SPI only the sum of read_len and write_len matters. The
behavior is the same no matter how the transfer length is divided behavior is the same no matter how the total transfer length is
between the two. For 3-wire SPI, the direction of mosi/miso is divided between the two. For 3-wire SPI, the direction of mosi/miso
switched from output to input after write_len cycles, at the is switched from output to input after write_len cycles, at the
"output" clk edge corresponding to bit write_len + 1 of the transfer. "shift_out" clk edge corresponding to bit write_length + 1 of the
transfer.
* Data output on mosi in 4-wire SPI during the read cycles is what * Data output on mosi in 4-wire SPI during the read cycles is what
is found in the data register at the time. is found in the data register at the time.
Data in the data register outside the least/most (depending Data in the data register outside the least/most (depending
on config.lsb_first) significant read_len bits is what is on config.lsb_first) significant read_length bits is what is
seen on miso during the write cycles. seen on miso during the write cycles.
* When the transfer is complete the wishbone transaction is ack-ed. * The SPI data register is double-buffered: once a transfer completes,
* Input data from the last transaction can be read from the data the previous transfer's read data is available in the data register
register at any time. and new write data can be written, queuing a new transfer. Transfers
submitted this way are chained and executed without deasserting cs.
* A wishbone transaction is ack-ed when the transfer has been written
to the intermediate buffer. It will be started when there are no
other transactions being executed.
Transaction Sequence: Transaction Sequence:
* If desired, write the config register to set up the core. * If desired, write the config register to set up the core.
* If desired, write the xfer register to change lengths and cs_n. * If desired, write the xfer register to change lengths and cs_n.
* Write the data register (also for zero-length writes), * Write the data register (also for zero-length writes),
writing triggers the transfer and when the transfer is complete the writing triggers the transfer and when the transfer is accepted to
write is ack-ed. the inermediate buffer, the write is ack-ed.
* If desired, read the data register. * If desired, read the data register.
* If desired write xfer and data for the next, chained, transfer
Register address and bit map: Register address and bit map:
config (address 2): config (address 2):
1 offline: all pins high-z (reset=1) 1 offline: all pins high-z (reset=1)
1 active: cs/transfer active
1 pending: transfer pending in intermediate buffer, bus writes will
block
1 cs_polarity: active level of chip select (reset=0) 1 cs_polarity: active level of chip select (reset=0)
1 clk_polarity: idle level for clk (reset=0) 1 clk_polarity: idle level for clk (reset=0)
1 clk_phase: first edge after cs assertion to sample data on (reset=0) 1 clk_phase: first edge after cs assertion to sample data on (reset=0)
@ -68,11 +226,11 @@ class SPIMaster(Module):
(1, 1): idle high, output on falling, input on rising (1, 1): idle high, output on falling, input on rising
1 lsb_first: LSB is the first bit on the wire (reset=0) 1 lsb_first: LSB is the first bit on the wire (reset=0)
1 half_duplex: 3-wire SPI, in/out on mosi (reset=0) 1 half_duplex: 3-wire SPI, in/out on mosi (reset=0)
10 undefined 12 div_write: counter load value to divide this module's clock
16 clk_load: clock load value to divide from this module's clock to the SPI write clk. clk pulses are asymmetric
to the SPI write clk clk pulses are asymmetric if the value is odd, favoring longer setup over hold times.
if a divider is odd, favoring longer setup over hold times. f_clk/f_spi_write == div_write + 2 (reset=0)
clk/spi_clk == clk_load + 2 (reset=0) 12 div_read: ditto for the read clock
xfer (address 1): xfer (address 1):
16 cs: active high bit mask of chip selects to assert 16 cs: active high bit mask of chip selects to assert
@ -89,50 +247,18 @@ class SPIMaster(Module):
### ###
# State machine
wb_we = Signal()
start = Signal()
active = Signal()
fsm = FSM("IDLE")
self.submodules += fsm
fsm.act("IDLE",
If(bus.cyc & bus.stb,
NextState("ACK"),
If(bus.we,
wb_we.eq(1),
If(bus.adr == 0, # data register
NextState("START"),
)
)
)
)
fsm.act("START",
start.eq(1),
NextState("ACTIVE"),
)
fsm.act("ACTIVE",
If(~active,
bus.ack.eq(1),
NextState("IDLE"),
)
)
fsm.act("ACK",
bus.ack.eq(1),
NextState("IDLE"),
)
# Wishbone # Wishbone
config = Record([ config = Record([
("offline", 1), ("offline", 1),
("active", 1),
("pending", 1),
("cs_polarity", 1), ("cs_polarity", 1),
("clk_polarity", 1), ("clk_polarity", 1),
("clk_phase", 1), ("clk_phase", 1),
("lsb_first", 1), ("lsb_first", 1),
("half_duplex", 1), ("half_duplex", 1),
("padding", 10), ("div_write", 12),
("clk_load", 16), ("div_read", 12),
]) ])
config.offline.reset = 1 config.offline.reset = 1
assert len(config) <= len(bus.dat_w) assert len(config) <= len(bus.dat_w)
@ -144,101 +270,96 @@ class SPIMaster(Module):
]) ])
assert len(xfer) <= len(bus.dat_w) assert len(xfer) <= len(bus.dat_w)
data = Signal.like(bus.dat_w)
wb_data = Array([data, xfer.raw_bits(), config.raw_bits()])[bus.adr]
self.comb += bus.dat_r.eq(wb_data)
self.sync += If(wb_we, wb_data.eq(bus.dat_w))
# SPI # SPI
write_count = Signal.like(xfer.write_length) spi = SPIMachine(data_width, clock_width=len(config.div_read),
read_count = Signal.like(xfer.read_length) bits_width=len(xfer.read_length))
clk_count = Signal.like(config.clk_load) self.submodules += spi
clk = Signal(reset=1) # idle high
phase = Signal() wb_we = Signal()
edge = Signal() pending = Signal()
write = Signal() cs = Signal.like(xfer.cs)
read = Signal() data_read = Signal.like(spi.reg.data)
miso = Signal() data_write = Signal.like(spi.reg.data)
miso_i = Signal()
mosi_o = Signal()
self.comb += [ self.comb += [
phase.eq(clk ^ config.clk_phase), wb_we.eq(bus.cyc & bus.stb & bus.we & (~pending | spi.done)),
edge.eq(active & (clk_count == 0)), bus.dat_r.eq(
write.eq(write_count != 0), Array([data_read, xfer.raw_bits(), config.raw_bits()
read.eq(read_count != 0), ])[bus.adr]),
spi.start.eq(pending & (~spi.cs | spi.done)),
spi.clk_phase.eq(config.clk_phase),
spi.reg.lsb.eq(config.lsb_first),
spi.div_write.eq(config.div_write),
spi.div_read.eq(config.div_read),
] ]
self.sync += [ self.sync += [
If(start, bus.ack.eq(~bus.we | ~pending | spi.done),
write_count.eq(xfer.write_length), If(wb_we,
read_count.eq(xfer.read_length), Array([data_write, xfer.raw_bits(), config.raw_bits()
active.eq(1), ])[bus.adr].eq(bus.dat_w)
), ),
If(active, config.active.eq(spi.cs),
clk_count.eq(clk_count - 1), config.pending.eq(pending),
If(spi.done,
data_read.eq(spi.reg.data),
), ),
If(start | edge, If(spi.start,
# setup time passes during phase 0 cs.eq(xfer.cs),
# use the lsb to bias that time to favor longer setup times spi.bits.n_write.eq(xfer.write_length),
clk_count.eq(config.clk_load[1:] + spi.bits.n_read.eq(xfer.read_length),
(config.clk_load[0] & phase)), spi.reg.data.eq(data_write),
clk.eq(~clk), # idle high pending.eq(0),
If(phase,
data.eq(Mux(config.lsb_first,
Cat(data[1:], miso),
Cat(miso, data[:-1]))),
mosi_o.eq(Mux(config.lsb_first, data[0], data[-1])),
If(write,
write_count.eq(write_count - 1),
), ),
).Else( If(wb_we & (bus.adr == 0), # data register
miso.eq(miso_i), pending.eq(1),
If(~write & read,
read_count.eq(read_count - 1),
),
),
),
If(~clk & edge & ~write & ~read, # always from low clk
active.eq(0),
), ),
] ]
# I/O # I/O
if hasattr(pads, "cs_n"):
cs_n_t = TSTriple(len(pads.cs_n)) cs_n_t = TSTriple(len(pads.cs_n))
self.specials += cs_n_t.get_tristate(pads.cs_n) self.specials += cs_n_t.get_tristate(pads.cs_n)
self.comb += [
cs_n_t.oe.eq(~config.offline),
cs_n_t.o.eq((cs & Replicate(spi.cs, len(cs))) ^
Replicate(~config.cs_polarity, len(cs))),
]
clk_t = TSTriple() clk_t = TSTriple()
self.specials += clk_t.get_tristate(pads.clk) self.specials += clk_t.get_tristate(pads.clk)
mosi_t = TSTriple() mosi_t = TSTriple()
self.specials += mosi_t.get_tristate(pads.mosi) self.specials += mosi_t.get_tristate(pads.mosi)
self.comb += [ self.comb += [
cs_n_t.oe.eq(~config.offline),
clk_t.oe.eq(~config.offline), clk_t.oe.eq(~config.offline),
mosi_t.oe.eq(~config.offline & (write | ~config.half_duplex)), mosi_t.oe.eq(~config.offline & spi.cs &
cs_n_t.o.eq((xfer.cs & Replicate(active, len(xfer.cs))) ^ (spi.oe | ~config.half_duplex)),
Replicate(~config.cs_polarity, len(xfer.cs))), clk_t.o.eq((spi.cg.clk & spi.cs) ^ config.clk_polarity),
clk_t.o.eq((clk & active) ^ config.clk_polarity), spi.reg.i.eq(Mux(config.half_duplex, mosi_t.i,
miso_i.eq(Mux(config.half_duplex, mosi_t.i,
getattr(pads, "miso", mosi_t.i))), getattr(pads, "miso", mosi_t.i))),
mosi_t.o.eq(mosi_o), mosi_t.o.eq(spi.reg.o),
] ]
SPI_CONFIG_ADDR = 2 SPI_DATA_ADDR, SPI_XFER_ADDR, SPI_CONFIG_ADDR = range(3)
SPI_XFER_ADDR = 1 (
SPI_DATA_ADDR = 0 SPI_OFFLINE,
SPI_OFFLINE = 1 << 0 SPI_ACTIVE,
SPI_CS_POLARITY = 1 << 1 SPI_PENDING,
SPI_CLK_POLARITY = 1 << 2 SPI_CS_POLARITY,
SPI_CLK_PHASE = 1 << 3 SPI_CLK_POLARITY,
SPI_LSB_FIRST = 1 << 4 SPI_CLK_PHASE,
SPI_HALF_DUPLEX = 1 << 5 SPI_LSB_FIRST,
SPI_HALF_DUPLEX,
) = (1 << i for i in range(8))
def SPI_CLK_LOAD(i): def SPI_DIV_WRITE(i):
return i << 16 return i << 8
def SPI_DIV_READ(i):
return i << 20
def SPI_CS(i): def SPI_CS(i):
@ -253,53 +374,44 @@ def SPI_READ_LENGTH(i):
return i << 24 return i << 24
def _test_gen(bus): def _test_xfer(bus, cs, wlen, rlen, wdata):
yield from bus.write(SPI_CONFIG_ADDR, yield from bus.write(SPI_XFER_ADDR, SPI_CS(cs) |
1*SPI_CLK_PHASE | 0*SPI_LSB_FIRST | SPI_WRITE_LENGTH(wlen) | SPI_READ_LENGTH(rlen))
1*SPI_HALF_DUPLEX | SPI_CLK_LOAD(3)) yield from bus.write(SPI_DATA_ADDR, wdata)
yield
yield from bus.write(SPI_XFER_ADDR, SPI_CS(0b00001) |
SPI_WRITE_LENGTH(4) | SPI_READ_LENGTH(0))
yield
yield from bus.write(SPI_DATA_ADDR, 0x90000000)
yield
print(hex((yield from bus.read(SPI_DATA_ADDR))))
yield
yield from bus.write(SPI_XFER_ADDR, SPI_CS(0b00010) |
SPI_WRITE_LENGTH(4) | SPI_READ_LENGTH(4))
yield
yield from bus.write(SPI_DATA_ADDR, 0x81000000)
yield
print(hex((yield from bus.read(SPI_DATA_ADDR))))
yield
yield from bus.write(SPI_XFER_ADDR, SPI_CS(0b00010) |
SPI_WRITE_LENGTH(0) | SPI_READ_LENGTH(4))
yield
yield from bus.write(SPI_DATA_ADDR, 0x90000000)
yield
print(hex((yield from bus.read(SPI_DATA_ADDR))))
yield
yield from bus.write(SPI_XFER_ADDR, SPI_CS(0b00010) |
SPI_WRITE_LENGTH(32) | SPI_READ_LENGTH(0))
yield
yield from bus.write(SPI_DATA_ADDR, 0x87654321)
yield
print(hex((yield from bus.read(SPI_DATA_ADDR))))
yield yield
def _test_read(bus, sync=SPI_ACTIVE | SPI_PENDING):
while (yield from bus.read(SPI_CONFIG_ADDR)) & sync:
pass
return (yield from bus.read(SPI_DATA_ADDR))
def _test_gen(bus):
yield from bus.write(SPI_CONFIG_ADDR,
0*SPI_CLK_PHASE | 0*SPI_LSB_FIRST |
1*SPI_HALF_DUPLEX |
SPI_DIV_WRITE(3) | SPI_DIV_READ(5))
yield from _test_xfer(bus, 0b01, 4, 0, 0x90000000)
print(hex((yield from _test_read(bus))))
yield from _test_xfer(bus, 0b10, 0, 4, 0x90000000)
print(hex((yield from _test_read(bus))))
yield from _test_xfer(bus, 0b11, 4, 4, 0x81000000)
print(hex((yield from _test_read(bus))))
yield from _test_xfer(bus, 0b01, 8, 32, 0x87654321)
yield from _test_xfer(bus, 0b01, 0, 32, 0x12345678)
print(hex((yield from _test_read(bus, SPI_PENDING))))
print(hex((yield from _test_read(bus, SPI_ACTIVE))))
return return
for cpol, cpha, lsb, clk in product( for cpol, cpha, lsb, clk in product(
(0, 1), (0, 1), (0, 1), (0, 1)): (0, 1), (0, 1), (0, 1), (0, 1)):
yield from bus.write(SPI_CONFIG_ADDR, yield from bus.write(SPI_CONFIG_ADDR,
cpol*SPI_CLK_POLARITY | cpha*SPI_CLK_PHASE | cpol*SPI_CLK_POLARITY | cpha*SPI_CLK_PHASE |
lsb*SPI_LSB_FIRST | SPI_CLK_LOAD(clk)) lsb*SPI_LSB_FIRST | SPI_DIV_WRITE(clk) |
SPI_DIV_READ(clk))
for wlen, rlen, wdata in product((0, 8, 32), (0, 8, 32), for wlen, rlen, wdata in product((0, 8, 32), (0, 8, 32),
(0, 0xffffffff, 0xdeadbeef)): (0, 0xffffffff, 0xdeadbeef)):
yield from bus.write(SPI_XFER_ADDR, SPI_CS(0b00001) | rdata = (yield from _test_xfer(bus, 0b1, wlen, rlen, wdata, True))
SPI_WRITE_LENGTH(wlen) |
SPI_READ_LENGTH(rlen))
yield from bus.write(SPI_DATA_ADDR, wdata)
rdata = yield from bus.read(SPI_DATA_ADDR)
len = (wlen + rlen) % 32 len = (wlen + rlen) % 32
mask = (1 << len) - 1 mask = (1 << len) - 1
if lsb: if lsb:
@ -336,10 +448,10 @@ if __name__ == "__main__":
] ]
Tristate.lower = staticmethod(lambda dr: T(dr)) Tristate.lower = staticmethod(lambda dr: T(dr))
from migen.fhdl.verilog import convert
pads = _TestPads() pads = _TestPads()
dut = SPIMaster(pads) dut = SPIMaster(pads)
dut.comb += pads.miso.eq(pads.mosi) dut.comb += pads.miso.eq(pads.mosi)
# from migen.fhdl.verilog import convert
# print(convert(dut)) # print(convert(dut))
run_simulation(dut, _test_gen(dut.bus), vcd_name="spi_master.vcd") run_simulation(dut, _test_gen(dut.bus), vcd_name="spi_master.vcd")