1# Copyright (c) 2006-2007 The Regents of The University of Michigan 2# Copyright (c) 2009,2015 Advanced Micro Devices, Inc. 3# Copyright (c) 2013 Mark D. Hill and David A. Wood 4# All rights reserved. 5# 6# Redistribution and use in source and binary forms, with or without 7# modification, are permitted provided that the following conditions are 8# met: redistributions of source code must retain the above copyright 9# notice, this list of conditions and the following disclaimer; 10# redistributions in binary form must reproduce the above copyright 11# notice, this list of conditions and the following disclaimer in the 12# documentation and/or other materials provided with the distribution; 13# neither the name of the copyright holders nor the names of its 14# contributors may be used to endorse or promote products derived from 15# this software without specific prior written permission. 16# 17# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 18# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 19# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 20# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 21# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 22# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 23# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 24# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 25# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 26# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 27# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 28# 29# Authors: Brad Beckmann 30# Nilay Vaish 31 32import math 33import m5 34from m5.objects import * 35from m5.defines import buildEnv 36from Ruby import create_topology, create_directories 37from Ruby import send_evicts 38from common import FileSystemConfig 39 40# 41# Declare caches used by the protocol 42# 43class L0Cache(RubyCache): pass 44class L1Cache(RubyCache): pass 45class L2Cache(RubyCache): pass 46 47def define_options(parser): 48 parser.add_option("--num-clusters", type = "int", default = 1, 49 help = "number of clusters in a design in which there are shared\ 50 caches private to clusters") 51 return 52 53def create_system(options, full_system, system, dma_ports, bootmem, 54 ruby_system): 55 56 if buildEnv['PROTOCOL'] != 'MESI_Three_Level': 57 fatal("This script requires the MESI_Three_Level protocol to be\ 58 built.") 59 60 cpu_sequencers = [] 61 62 # 63 # The ruby network creation expects the list of nodes in the system to be 64 # consistent with the NetDest list. Therefore the l1 controller nodes 65 # must be listed before the directory nodes and directory nodes before 66 # dma nodes, etc. 67 # 68 l0_cntrl_nodes = [] 69 l1_cntrl_nodes = [] 70 l2_cntrl_nodes = [] 71 dma_cntrl_nodes = [] 72 73 assert (options.num_cpus % options.num_clusters == 0) 74 num_cpus_per_cluster = options.num_cpus / options.num_clusters 75 76 assert (options.num_l2caches % options.num_clusters == 0) 77 num_l2caches_per_cluster = options.num_l2caches / options.num_clusters 78 79 l2_bits = int(math.log(num_l2caches_per_cluster, 2)) 80 block_size_bits = int(math.log(options.cacheline_size, 2)) 81 l2_index_start = block_size_bits + l2_bits 82 83 # 84 # Must create the individual controllers before the network to ensure the 85 # controller constructors are called before the network constructor 86 # 87 for i in range(options.num_clusters): 88 for j in range(num_cpus_per_cluster): 89 # 90 # First create the Ruby objects associated with this cpu 91 # 92 l0i_cache = L0Cache(size = '4096B', assoc = 1, is_icache = True, 93 start_index_bit = block_size_bits, 94 replacement_policy = LRUReplacementPolicy()) 95 96 l0d_cache = L0Cache(size = '4096B', assoc = 1, is_icache = False, 97 start_index_bit = block_size_bits, 98 replacement_policy = LRUReplacementPolicy()) 99 100 # the ruby random tester reuses num_cpus to specify the 101 # number of cpu ports connected to the tester object, which 102 # is stored in system.cpu. because there is only ever one 103 # tester object, num_cpus is not necessarily equal to the 104 # size of system.cpu; therefore if len(system.cpu) == 1 105 # we use system.cpu[0] to set the clk_domain, thereby ensuring 106 # we don't index off the end of the cpu list. 107 if len(system.cpu) == 1: 108 clk_domain = system.cpu[0].clk_domain 109 else: 110 clk_domain = system.cpu[i].clk_domain 111 112 l0_cntrl = L0Cache_Controller( 113 version = i * num_cpus_per_cluster + j, Icache = l0i_cache, 114 Dcache = l0d_cache, send_evictions = send_evicts(options), 115 clk_domain = clk_domain, ruby_system = ruby_system) 116 117 cpu_seq = RubySequencer(version = i * num_cpus_per_cluster + j, 118 icache = l0i_cache, 119 clk_domain = clk_domain, 120 dcache = l0d_cache, 121 ruby_system = ruby_system) 122 123 l0_cntrl.sequencer = cpu_seq 124 125 l1_cache = L1Cache(size = options.l1d_size, 126 assoc = options.l1d_assoc, 127 start_index_bit = block_size_bits, 128 is_icache = False) 129 130 l1_cntrl = L1Cache_Controller( 131 version = i * num_cpus_per_cluster + j, 132 cache = l1_cache, l2_select_num_bits = l2_bits, 133 cluster_id = i, ruby_system = ruby_system) 134 135 exec("ruby_system.l0_cntrl%d = l0_cntrl" 136 % ( i * num_cpus_per_cluster + j)) 137 exec("ruby_system.l1_cntrl%d = l1_cntrl" 138 % ( i * num_cpus_per_cluster + j)) 139 140 # 141 # Add controllers and sequencers to the appropriate lists 142 # 143 cpu_sequencers.append(cpu_seq) 144 l0_cntrl_nodes.append(l0_cntrl) 145 l1_cntrl_nodes.append(l1_cntrl) 146 147 # Connect the L0 and L1 controllers 148 l0_cntrl.mandatoryQueue = MessageBuffer() 149 l0_cntrl.bufferToL1 = MessageBuffer(ordered = True) 150 l1_cntrl.bufferFromL0 = l0_cntrl.bufferToL1 151 l0_cntrl.bufferFromL1 = MessageBuffer(ordered = True) 152 l1_cntrl.bufferToL0 = l0_cntrl.bufferFromL1 153 154 # Connect the L1 controllers and the network 155 l1_cntrl.requestToL2 = MessageBuffer() 156 l1_cntrl.requestToL2.master = ruby_system.network.slave 157 l1_cntrl.responseToL2 = MessageBuffer() 158 l1_cntrl.responseToL2.master = ruby_system.network.slave 159 l1_cntrl.unblockToL2 = MessageBuffer() 160 l1_cntrl.unblockToL2.master = ruby_system.network.slave 161 162 l1_cntrl.requestFromL2 = MessageBuffer() 163 l1_cntrl.requestFromL2.slave = ruby_system.network.master 164 l1_cntrl.responseFromL2 = MessageBuffer() 165 l1_cntrl.responseFromL2.slave = ruby_system.network.master 166 167 168 for j in range(num_l2caches_per_cluster): 169 l2_cache = L2Cache(size = options.l2_size, 170 assoc = options.l2_assoc, 171 start_index_bit = l2_index_start) 172 173 l2_cntrl = L2Cache_Controller( 174 version = i * num_l2caches_per_cluster + j, 175 L2cache = l2_cache, cluster_id = i, 176 transitions_per_cycle = options.ports, 177 ruby_system = ruby_system) 178 179 exec("ruby_system.l2_cntrl%d = l2_cntrl" 180 % (i * num_l2caches_per_cluster + j)) 181 l2_cntrl_nodes.append(l2_cntrl) 182 183 # Connect the L2 controllers and the network 184 l2_cntrl.DirRequestFromL2Cache = MessageBuffer() 185 l2_cntrl.DirRequestFromL2Cache.master = ruby_system.network.slave 186 l2_cntrl.L1RequestFromL2Cache = MessageBuffer() 187 l2_cntrl.L1RequestFromL2Cache.master = ruby_system.network.slave 188 l2_cntrl.responseFromL2Cache = MessageBuffer() 189 l2_cntrl.responseFromL2Cache.master = ruby_system.network.slave 190 191 l2_cntrl.unblockToL2Cache = MessageBuffer() 192 l2_cntrl.unblockToL2Cache.slave = ruby_system.network.master 193 l2_cntrl.L1RequestToL2Cache = MessageBuffer() 194 l2_cntrl.L1RequestToL2Cache.slave = ruby_system.network.master 195 l2_cntrl.responseToL2Cache = MessageBuffer() 196 l2_cntrl.responseToL2Cache.slave = ruby_system.network.master 197 198 # Run each of the ruby memory controllers at a ratio of the frequency of 199 # the ruby system 200 # clk_divider value is a fix to pass regression. 201 ruby_system.memctrl_clk_domain = DerivedClockDomain( 202 clk_domain = ruby_system.clk_domain, clk_divider = 3) 203 204 mem_dir_cntrl_nodes, rom_dir_cntrl_node = create_directories( 205 options, bootmem, ruby_system, system) 206 dir_cntrl_nodes = mem_dir_cntrl_nodes[:] 207 if rom_dir_cntrl_node is not None: 208 dir_cntrl_nodes.append(rom_dir_cntrl_node) 209 for dir_cntrl in dir_cntrl_nodes: 210 # Connect the directory controllers and the network 211 dir_cntrl.requestToDir = MessageBuffer() 212 dir_cntrl.requestToDir.slave = ruby_system.network.master 213 dir_cntrl.responseToDir = MessageBuffer() 214 dir_cntrl.responseToDir.slave = ruby_system.network.master 215 dir_cntrl.responseFromDir = MessageBuffer() 216 dir_cntrl.responseFromDir.master = ruby_system.network.slave 217 dir_cntrl.responseFromMemory = MessageBuffer() 218 219 for i, dma_port in enumerate(dma_ports): 220 # 221 # Create the Ruby objects associated with the dma controller 222 # 223 dma_seq = DMASequencer(version = i, ruby_system = ruby_system) 224 225 dma_cntrl = DMA_Controller(version = i, 226 dma_sequencer = dma_seq, 227 transitions_per_cycle = options.ports, 228 ruby_system = ruby_system) 229 230 exec("ruby_system.dma_cntrl%d = dma_cntrl" % i) 231 exec("ruby_system.dma_cntrl%d.dma_sequencer.slave = dma_port" % i) 232 dma_cntrl_nodes.append(dma_cntrl) 233 234 # Connect the dma controller to the network 235 dma_cntrl.mandatoryQueue = MessageBuffer() 236 dma_cntrl.responseFromDir = MessageBuffer(ordered = True) 237 dma_cntrl.responseFromDir.slave = ruby_system.network.master 238 dma_cntrl.requestToDir = MessageBuffer() 239 dma_cntrl.requestToDir.master = ruby_system.network.slave 240 241 all_cntrls = l0_cntrl_nodes + \ 242 l1_cntrl_nodes + \ 243 l2_cntrl_nodes + \ 244 dir_cntrl_nodes + \ 245 dma_cntrl_nodes 246 247 # Create the io controller and the sequencer 248 if full_system: 249 io_seq = DMASequencer(version=len(dma_ports), ruby_system=ruby_system) 250 ruby_system._io_port = io_seq 251 io_controller = DMA_Controller(version = len(dma_ports), 252 dma_sequencer = io_seq, 253 ruby_system = ruby_system) 254 ruby_system.io_controller = io_controller 255 256 # Connect the dma controller to the network 257 io_controller.mandatoryQueue = MessageBuffer() 258 io_controller.responseFromDir = MessageBuffer(ordered = True) 259 io_controller.responseFromDir.slave = ruby_system.network.master 260 io_controller.requestToDir = MessageBuffer() 261 io_controller.requestToDir.master = ruby_system.network.slave 262 263 all_cntrls = all_cntrls + [io_controller] 264 # Register configuration with filesystem 265 else: 266 for i in xrange(options.num_clusters): 267 for j in xrange(num_cpus_per_cluster): 268 FileSystemConfig.register_cpu(physical_package_id = 0, 269 core_siblings = xrange(options.num_cpus), 270 core_id = i*num_cpus_per_cluster+j, 271 thread_siblings = []) 272 273 FileSystemConfig.register_cache(level = 0, 274 idu_type = 'Instruction', 275 size = '4096B', 276 line_size = options.cacheline_size, 277 assoc = 1, 278 cpus = [i*num_cpus_per_cluster+j]) 279 FileSystemConfig.register_cache(level = 0, 280 idu_type = 'Data', 281 size = '4096B', 282 line_size = options.cacheline_size, 283 assoc = 1, 284 cpus = [i*num_cpus_per_cluster+j]) 285 286 FileSystemConfig.register_cache(level = 1, 287 idu_type = 'Unified', 288 size = options.l1d_size, 289 line_size = options.cacheline_size, 290 assoc = options.l1d_assoc, 291 cpus = [i*num_cpus_per_cluster+j]) 292 293 FileSystemConfig.register_cache(level = 2, 294 idu_type = 'Unified', 295 size = str(MemorySize(options.l2_size) * \ 296 num_l2caches_per_cluster)+'B', 297 line_size = options.cacheline_size, 298 assoc = options.l2_assoc, 299 cpus = [n for n in xrange(i*num_cpus_per_cluster, \ 300 (i+1)*num_cpus_per_cluster)]) 301 302 ruby_system.network.number_of_virtual_networks = 3 303 topology = create_topology(all_cntrls, options) 304 return (cpu_sequencers, mem_dir_cntrl_nodes, topology) 305