Ruby.py revision 12065
11689SN/A# Copyright (c) 2012, 2017 ARM Limited
22329SN/A# All rights reserved.
31689SN/A#
41689SN/A# The license below extends only to copyright in the software and shall
51689SN/A# not be construed as granting a license to any other intellectual
61689SN/A# property including but not limited to intellectual property relating
71689SN/A# to a hardware implementation of the functionality of the software
81689SN/A# licensed hereunder.  You may use the software subject to the license
91689SN/A# terms below provided that you ensure that this notice is replicated
101689SN/A# unmodified and in its entirety in all distributions of the software,
111689SN/A# modified or unmodified, in source code or in binary form.
121689SN/A#
131689SN/A# Copyright (c) 2006-2007 The Regents of The University of Michigan
141689SN/A# Copyright (c) 2009 Advanced Micro Devices, Inc.
151689SN/A# All rights reserved.
161689SN/A#
171689SN/A# Redistribution and use in source and binary forms, with or without
181689SN/A# modification, are permitted provided that the following conditions are
191689SN/A# met: redistributions of source code must retain the above copyright
201689SN/A# notice, this list of conditions and the following disclaimer;
211689SN/A# redistributions in binary form must reproduce the above copyright
221689SN/A# notice, this list of conditions and the following disclaimer in the
231689SN/A# documentation and/or other materials provided with the distribution;
241689SN/A# neither the name of the copyright holders nor the names of its
251689SN/A# contributors may be used to endorse or promote products derived from
261689SN/A# this software without specific prior written permission.
272665Ssaidi@eecs.umich.edu#
282665Ssaidi@eecs.umich.edu# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
291689SN/A# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
301689SN/A# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
311858SN/A# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
321717SN/A# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
331060SN/A# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
342292SN/A# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
352292SN/A# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
361061SN/A# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
372292SN/A# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
382292SN/A# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
392292SN/A#
401060SN/A# Authors: Brad Beckmann
411060SN/A
421060SN/Aimport math
432292SN/Aimport m5
442292SN/Afrom m5.objects import *
451060SN/Afrom m5.defines import buildEnv
462292SN/Afrom m5.util import addToPath, fatal
472292SN/A
482292SN/Afrom common import MemConfig
492292SN/A
502292SN/Afrom topologies import *
512292SN/Afrom network import Network
522292SN/A
532292SN/Adef define_options(parser):
542292SN/A    # By default, ruby uses the simple timing cpu
552292SN/A    parser.set_defaults(cpu_type="TimingSimpleCPU")
562292SN/A
572292SN/A    parser.add_option("--ruby-clock", action="store", type="string",
582292SN/A                      default='2GHz',
592292SN/A                      help="Clock for blocks running at Ruby system's speed")
602292SN/A
612292SN/A    parser.add_option("--access-backing-store", action="store_true", default=False,
622292SN/A                      help="Should ruby maintain a second copy of memory")
632292SN/A
642292SN/A    # Options related to cache structure
652292SN/A    parser.add_option("--ports", action="store", type="int", default=4,
662292SN/A                      help="used of transitions per cycle which is a proxy \
672292SN/A                            for the number of ports.")
682292SN/A
692292SN/A    # network options are in network/Network.py
702292SN/A
712292SN/A    # ruby mapping options
722292SN/A    parser.add_option("--numa-high-bit", type="int", default=0,
732292SN/A                      help="high order address bit to use for numa mapping. " \
742292SN/A                           "0 = highest bit, not specified = lowest bit")
752292SN/A
762292SN/A    parser.add_option("--recycle-latency", type="int", default=10,
772292SN/A                      help="Recycle latency for ruby controller input buffers")
782292SN/A
792292SN/A    protocol = buildEnv['PROTOCOL']
802292SN/A    exec "import %s" % protocol
812292SN/A    eval("%s.define_options(parser)" % protocol)
822292SN/A    Network.define_options(parser)
832292SN/A
842292SN/Adef setup_memory_controllers(system, ruby, dir_cntrls, options):
852292SN/A    ruby.block_size_bytes = options.cacheline_size
862292SN/A    ruby.memory_size_bits = 48
872292SN/A
882292SN/A    index = 0
892292SN/A    mem_ctrls = []
902292SN/A    crossbars = []
912292SN/A
922292SN/A    # Sets bits to be used for interleaving.  Creates memory controllers
932292SN/A    # attached to a directory controller.  A separate controller is created
942292SN/A    # for each address range as the abstract memory can handle only one
952292SN/A    # contiguous address range as of now.
962292SN/A    for dir_cntrl in dir_cntrls:
972292SN/A        crossbar = None
982292SN/A        if len(system.mem_ranges) > 1:
991060SN/A            crossbar = IOXBar()
1001060SN/A            crossbars.append(crossbar)
1011061SN/A            dir_cntrl.memory = crossbar.slave
1021060SN/A
1031060SN/A        for r in system.mem_ranges:
1041060SN/A            mem_ctrl = MemConfig.create_mem_ctrl(
1051060SN/A                MemConfig.get(options.mem_type), r, index, options.num_dirs,
1061060SN/A                int(math.log(options.num_dirs, 2)), options.cacheline_size)
1072292SN/A
1082292SN/A            if options.access_backing_store:
1092292SN/A                mem_ctrl.kvm_map=False
1102292SN/A
1111060SN/A            mem_ctrls.append(mem_ctrl)
1122292SN/A
1132292SN/A            if crossbar != None:
1142292SN/A                mem_ctrl.port = crossbar.master
1152292SN/A            else:
1162292SN/A                mem_ctrl.port = dir_cntrl.memory
1172292SN/A
1182292SN/A        index += 1
1192292SN/A
1202292SN/A    system.mem_ctrls = mem_ctrls
1212292SN/A
1222292SN/A    if len(crossbars) > 0:
1232292SN/A        ruby.crossbars = crossbars
1242292SN/A
1252292SN/A
1262307SN/Adef create_topology(controllers, options):
1272307SN/A    """ Called from create_system in configs/ruby/<protocol>.py
1282307SN/A        Must return an object which is a subclass of BaseTopology
1292307SN/A        found in configs/topologies/BaseTopology.py
1302307SN/A        This is a wrapper for the legacy topologies.
1312307SN/A    """
1322307SN/A    exec "import topologies.%s as Topo" % options.topology
1332307SN/A    topology = eval("Topo.%s(controllers)" % options.topology)
1342307SN/A    return topology
1352307SN/A
1362307SN/Adef create_system(options, full_system, system, piobus = None, dma_ports = []):
1372307SN/A
1382307SN/A    system.ruby = RubySystem()
1392307SN/A    ruby = system.ruby
1402307SN/A
1412307SN/A    # Create the network object
1422307SN/A    (network, IntLinkClass, ExtLinkClass, RouterClass, InterfaceClass) = \
1432307SN/A        Network.create_network(options, ruby)
1442307SN/A    ruby.network = network
1452307SN/A
1462307SN/A    protocol = buildEnv['PROTOCOL']
1472307SN/A    exec "import %s" % protocol
1482307SN/A    try:
1492307SN/A        (cpu_sequencers, dir_cntrls, topology) = \
1502307SN/A             eval("%s.create_system(options, full_system, system, dma_ports,\
1512292SN/A                                    ruby)"
1522292SN/A                  % protocol)
1532292SN/A    except:
1542292SN/A        print "Error: could not create sytem for ruby protocol %s" % protocol
1552292SN/A        raise
1562292SN/A
1572292SN/A    # Create the network topology
1582292SN/A    topology.makeTopology(options, network, IntLinkClass, ExtLinkClass,
1592292SN/A            RouterClass)
1602292SN/A
1612292SN/A    # Initialize network based on topology
1622292SN/A    Network.init_network(options, network, InterfaceClass)
1632292SN/A
1642292SN/A    # Create a port proxy for connecting the system port. This is
1652292SN/A    # independent of the protocol and kept in the protocol-agnostic
1662292SN/A    # part (i.e. here).
1672292SN/A    sys_port_proxy = RubyPortProxy(ruby_system = ruby)
1682292SN/A    if piobus is not None:
1692292SN/A        sys_port_proxy.pio_master_port = piobus.slave
1702292SN/A
1712292SN/A    # Give the system port proxy a SimObject parent without creating a
1722292SN/A    # full-fledged controller
1732292SN/A    system.sys_port_proxy = sys_port_proxy
1742292SN/A
1752292SN/A    # Connect the system port for loading of binaries etc
1762292SN/A    system.system_port = system.sys_port_proxy.slave
1772292SN/A
1782292SN/A    setup_memory_controllers(system, ruby, dir_cntrls, options)
1792292SN/A
1802292SN/A    # Connect the cpu sequencers and the piobus
1811060SN/A    if piobus != None:
1821060SN/A        for cpu_seq in cpu_sequencers:
1831061SN/A            cpu_seq.pio_master_port = piobus.slave
1841060SN/A            cpu_seq.mem_master_port = piobus.slave
1851060SN/A
1861060SN/A            if buildEnv['TARGET_ISA'] == "x86":
1872292SN/A                cpu_seq.pio_slave_port = piobus.master
1881061SN/A
1892292SN/A    ruby.number_of_virtual_networks = ruby.network.number_of_virtual_networks
1902292SN/A    ruby._cpu_ports = cpu_sequencers
1911060SN/A    ruby.num_of_sequencers = len(cpu_sequencers)
1922292SN/A
1932292SN/A    # Create a backing copy of physical memory in case required
1941060SN/A    if options.access_backing_store:
1952292SN/A        ruby.access_backing_store = True
1962292SN/A        ruby.phys_mem = SimpleMemory(range=system.mem_ranges[0],
1972292SN/A                                     in_addr_map=False)
1982292SN/A
1992292SN/Adef create_directories(options, mem_ranges, ruby_system):
2001060SN/A    dir_cntrl_nodes = []
2011060SN/A    if options.numa_high_bit:
2021061SN/A        numa_bit = options.numa_high_bit
2031060SN/A    else:
2041061SN/A        # if the numa_bit is not specified, set the directory bits as the
2051060SN/A        # lowest bits above the block offset bits, and the numa_bit as the
2062292SN/A        # highest of those directory bits
2071060SN/A        dir_bits = int(math.log(options.num_dirs, 2))
2081060SN/A        block_size_bits = int(math.log(options.cacheline_size, 2))
2092292SN/A        numa_bit = block_size_bits + dir_bits - 1
2101060SN/A
2111060SN/A    for i in xrange(options.num_dirs):
2121060SN/A        dir_ranges = []
2132292SN/A        for r in mem_ranges:
2141060SN/A            addr_range = m5.objects.AddrRange(r.start, size = r.size(),
2152292SN/A                                              intlvHighBit = numa_bit,
2162292SN/A                                              intlvBits = dir_bits,
2172292SN/A                                              intlvMatch = i)
2182292SN/A            dir_ranges.append(addr_range)
2192292SN/A
2202292SN/A        dir_cntrl = Directory_Controller()
2211060SN/A        dir_cntrl.version = i
2221060SN/A        dir_cntrl.directory = RubyDirectoryMemory()
2232292SN/A        dir_cntrl.ruby_system = ruby_system
2242292SN/A        dir_cntrl.addr_ranges = dir_ranges
2252292SN/A
2262292SN/A        exec("ruby_system.dir_cntrl%d = dir_cntrl" % i)
2272292SN/A        dir_cntrl_nodes.append(dir_cntrl)
2282292SN/A    return dir_cntrl_nodes
2292292SN/A
2302292SN/Adef send_evicts(options):
2312292SN/A    # currently, 2 scenarios warrant forwarding evictions to the CPU:
2322292SN/A    # 1. The O3 model must keep the LSQ coherent with the caches
2331060SN/A    # 2. The x86 mwait instruction is built on top of coherence invalidations
2341060SN/A    if options.cpu_type == "DerivO3CPU" or buildEnv['TARGET_ISA'] == 'x86':
2352292SN/A        return True
2361060SN/A    return False
2371060SN/A