def createMemoryHierarchy(self):
self.createDMADevices()
# VIPER requires the number of instruction and scalar caches
if (buildEnv['PROTOCOL'] == 'GPU_VIPER'):
# Currently, the sqc (I-Cache of GPU) is shared by
# multiple compute units(CUs). The protocol works just fine
# even if sqc is not shared. Overriding this option here
# so that the user need not explicitly set this (assuming
# sharing sqc is the common usage)
self._opts.num_sqc = \
int(math.ceil(float(self._opts.num_compute_units)\
/ self._opts.cu_per_sqc))
self._opts.num_scalar_cache = \
int(math.ceil(float(self._opts.num_compute_units)\
/ self._opts.cu_per_scalar_cache))
Ruby.create_system(self._opts, True, self, self.iobus, self._dma_ports,
None)
# don't connect ide as it gets connected in attachIO call
for dma_port in self._dma_ports[1:]:
dma_port.pio = self.iobus.master
self.ruby.clk_domain = SrcClockDomain()
self.ruby.clk_domain.clock = self._opts.ruby_clock
self.ruby.clk_domain.voltage_domain = VoltageDomain()
for i, cpu in enumerate(self.cpu):
cpu.icache_port = self.ruby._cpu_ports[i].slave
cpu.dcache_port = self.ruby._cpu_ports[i].slave
cpu.itb.walker.port = self.ruby._cpu_ports[i].slave
cpu.dtb.walker.port = self.ruby._cpu_ports[i].slave
if self._opts.dgpu or self._opts.apu:
gpu_port_idx = len(self.ruby._cpu_ports) \
- self._opts.num_compute_units \
- self._opts.num_sqc \
- self._opts.num_scalar_cache
for i in xrange(self._opts.num_compute_units):
for j in xrange(self._opts.wf_size):
self.shader.CUs[i].memory_port[j] = \
self.ruby._cpu_ports[gpu_port_idx].slave[j]
gpu_port_idx += 1
for i in xrange(self._opts.num_compute_units):
if i > 0 and not i % self._opts.cu_per_sqc:
gpu_port_idx += 1
self.shader.CUs[i].sqc_port = \
self.ruby._cpu_ports[gpu_port_idx].slave
gpu_port_idx += 1
for i in xrange(self._opts.num_compute_units):
if i > 0 and not i % self._opts.cu_per_scalar_cache:
gpu_port_idx += 1
self.shader.CUs[i].scalar_port = \
self.ruby._cpu_ports[gpu_port_idx].slave
def config_ruby(system, args):
cpus = []
for cluster in system.cpu_cluster:
for cpu in cluster.cpus:
cpus.append(cpu)
Ruby.create_system(args, True, system, system.iobus, system._dma_ports,
system.realview.bootmem, cpus)
# Create a seperate clock domain for Ruby
system.ruby.clk_domain = SrcClockDomain(
clock=args.ruby_clock, voltage_domain=system.voltage_domain)
def createRubySystem(self):
bootmem = getattr(self, 'bootmem', None)
Ruby.create_system(self._opts, True, self, self.iobus, self._dma_ports,
bootmem)
self.iobus.master = self.ruby._io_port.slave
for (i, cpu) in enumerate(self.cpu):
# Tie the cpu ports to the correct ruby system ports
cpu.icache_port = self.ruby._cpu_ports[i].slave
cpu.dcache_port = self.ruby._cpu_ports[i].slave
# For x86/arm only
cpu.itb.walker.port = self.ruby._cpu_ports[i].slave
cpu.dtb.walker.port = self.ruby._cpu_ports[i].slave
def init_system(self, system):
"""Initialize a system.
Arguments:
system -- System to initialize.
"""
self.create_clk_src(system)
system.cpu = self.create_cpus(system.cpu_clk_domain)
if _have_kvm_support and \
any([isinstance(c, BaseKvmCPU) for c in system.cpu]):
self.init_kvm(system)
if self.use_ruby:
# Add the ruby specific and protocol specific options
parser = optparse.OptionParser()
Options.addCommonOptions(parser)
Ruby.define_options(parser)
(options, args) = parser.parse_args()
# Set the default cache size and associativity to be very
# small to encourage races between requests and writebacks.
options.l1d_size="32kB"
options.l1i_size="32kB"
options.l2_size="4MB"
options.l1d_assoc=4
options.l1i_assoc=2
options.l2_assoc=8
options.num_cpus = self.num_cpus
options.num_dirs = 2
bootmem = getattr(system, 'bootmem', None)
Ruby.create_system(options, True, system, system.iobus,
system._dma_ports, bootmem)
# Create a seperate clock domain for Ruby
system.ruby.clk_domain = SrcClockDomain(
clock = options.ruby_clock,
voltage_domain = system.voltage_domain)
for i, cpu in enumerate(system.cpu):
if not cpu.switched_out:
cpu.createInterruptController()
cpu.connectCachedPorts(system.ruby._cpu_ports[i])
else:
sha_bus = self.create_caches_shared(system)
for cpu in system.cpu:
self.init_cpu(system, cpu, sha_bus)
def init_system(self, system):
"""Initialize a system.
Arguments:
system -- System to initialize.
"""
self.create_clk_src(system)
system.cpu = self.create_cpus(system.cpu_clk_domain)
if self.use_ruby:
# Add the ruby specific and protocol specific options
parser = argparse.ArgumentParser()
Options.addCommonOptions(parser)
Ruby.define_options(parser)
args, extra = parser.parse_known_args()
# Set the default cache size and associativity to be very
# small to encourage races between requests and writebacks.
args.l1d_size = "32kB"
args.l1i_size = "32kB"
args.l2_size = "4MB"
args.l1d_assoc = 4
args.l1i_assoc = 2
args.l2_assoc = 8
args.num_cpus = self.num_cpus
args.num_dirs = 2
bootmem = getattr(system, '_bootmem', None)
Ruby.create_system(args, True, system, system.iobus,
system._dma_ports, bootmem)
# Create a seperate clock domain for Ruby
system.ruby.clk_domain = SrcClockDomain(
clock=args.ruby_clock, voltage_domain=system.voltage_domain)
for i, cpu in enumerate(system.cpu):
if not cpu.switched_out:
cpu.createInterruptController()
cpu.connectCachedPorts(system.ruby._cpu_ports[i].in_ports)
else:
sha_bus = self.create_caches_shared(system)
for cpu in system.cpu:
self.init_cpu(system, cpu, sha_bus)
if _have_kvm_support and \
any([isinstance(c, BaseKvmCPU) for c in system.cpu]):
self.init_kvm(system)
self.init_kvm_cpus(system.cpu)
Ruby=True)
system.kernel = SysPaths.binary('x86_64-vmlinux-2.6.22.9')
# Dummy voltage domain for all our clock domains
system.voltage_domain = VoltageDomain(voltage=options.sys_voltage)
system.kernel = FSConfig.binary('x86_64-vmlinux-2.6.22.9.smp')
system.clk_domain = SrcClockDomain(clock='1GHz',
voltage_domain=system.voltage_domain)
system.cpu_clk_domain = SrcClockDomain(clock='2GHz',
voltage_domain=system.voltage_domain)
system.cpu = [
TimingSimpleCPU(cpu_id=i, clk_domain=system.cpu_clk_domain)
for i in range(options.num_cpus)
]
Ruby.create_system(options, True, system, system.iobus, system._dma_ports)
# Create a seperate clock domain for Ruby
system.ruby.clk_domain = SrcClockDomain(clock=options.ruby_clock,
voltage_domain=system.voltage_domain)
# Connect the ruby io port to the PIO bus,
# assuming that there is just one such port.
system.iobus.master = system.ruby._io_port.slave
for (i, cpu) in enumerate(system.cpu):
# create the interrupt controller
cpu.createInterruptController()
# Tie the cpu ports to the correct ruby system ports
cpu.icache_port = system.ruby._cpu_ports[i].slave
cpu.dcache_port = system.ruby._cpu_ports[i].slave
def makeGpuFSSystem(args):
# Boot options are standard gem5 options plus:
# - Framebuffer device emulation 0 to reduce driver code paths.
# - Blacklist amdgpu as it cannot (currently) load in KVM CPU.
# - Blacklist psmouse as amdgpu driver adds proprietary commands that
# cause gem5 to panic.
boot_options = ['earlyprintk=ttyS0', 'console=ttyS0,9600',
'lpj=7999923', 'root=/dev/sda1',
'drm_kms_helper.fbdev_emulation=0',
'modprobe.blacklist=amdgpu',
'modprobe.blacklist=psmouse']
cmdline = ' '.join(boot_options)
if MemorySize(args.mem_size) < MemorySize('2GB'):
panic("Need at least 2GB of system memory to load amdgpu module")
# Use the common FSConfig to setup a Linux X86 System
(TestCPUClass, test_mem_mode, FutureClass) = Simulation.setCPUClass(args)
bm = SysConfig(disks=[args.disk_image], mem=args.mem_size)
system = makeLinuxX86System(test_mem_mode, args.num_cpus, bm, True,
cmdline=cmdline)
system.workload.object_file = binary(args.kernel)
# Set the cache line size for the entire system.
system.cache_line_size = args.cacheline_size
# Create a top-level voltage and clock domain.
system.voltage_domain = VoltageDomain(voltage = args.sys_voltage)
system.clk_domain = SrcClockDomain(clock = args.sys_clock,
voltage_domain = system.voltage_domain)
# Create a CPU voltage and clock domain.
system.cpu_voltage_domain = VoltageDomain()
system.cpu_clk_domain = SrcClockDomain(clock = args.cpu_clock,
voltage_domain =
system.cpu_voltage_domain)
# Setup VGA ROM region
system.shadow_rom_ranges = [AddrRange(0xc0000, size = Addr('128kB'))]
# Create specified number of CPUs. GPUFS really only needs one.
system.cpu = [TestCPUClass(clk_domain=system.cpu_clk_domain, cpu_id=i)
for i in range(args.num_cpus)]
if ObjectList.is_kvm_cpu(TestCPUClass) or \
ObjectList.is_kvm_cpu(FutureClass):
system.kvm_vm = KvmVM()
# Create AMDGPU and attach to southbridge
shader = createGPU(system, args)
connectGPU(system, args)
# This arbitrary address is something in the X86 I/O hole
hsapp_gpu_map_paddr = 0xe00000000
gpu_hsapp = HSAPacketProcessor(pioAddr=hsapp_gpu_map_paddr,
numHWQueues=args.num_hw_queues)
dispatcher = GPUDispatcher()
gpu_cmd_proc = GPUCommandProcessor(hsapp=gpu_hsapp,
dispatcher=dispatcher)
shader.dispatcher = dispatcher
shader.gpu_cmd_proc = gpu_cmd_proc
# GPU, HSAPP, and GPUCommandProc are DMA devices
system._dma_ports.append(gpu_hsapp)
system._dma_ports.append(gpu_cmd_proc)
system._dma_ports.append(system.pc.south_bridge.gpu)
gpu_hsapp.pio = system.iobus.mem_side_ports
gpu_cmd_proc.pio = system.iobus.mem_side_ports
system.pc.south_bridge.gpu.pio = system.iobus.mem_side_ports
# Create Ruby system using Ruby.py for now
Ruby.create_system(args, True, system, system.iobus,
system._dma_ports)
# Create a seperate clock domain for Ruby
system.ruby.clk_domain = SrcClockDomain(clock = args.ruby_clock,
voltage_domain = system.voltage_domain)
for (i, cpu) in enumerate(system.cpu):
#
# Tie the cpu ports to the correct ruby system ports
#
cpu.clk_domain = system.cpu_clk_domain
cpu.createThreads()
cpu.createInterruptController()
system.ruby._cpu_ports[i].connectCpuPorts(cpu)
# The shader core will be whatever is after the CPU cores are accounted for
shader_idx = args.num_cpus
system.cpu.append(shader)
gpu_port_idx = len(system.ruby._cpu_ports) \
- args.num_compute_units - args.num_sqc \
- args.num_scalar_cache
gpu_port_idx = gpu_port_idx - args.num_cp * 2
# Connect token ports. For this we need to search through the list of all
# sequencers, since the TCP coalescers will not necessarily be first. Only
# TCP coalescers use a token port for back pressure.
token_port_idx = 0
for i in range(len(system.ruby._cpu_ports)):
if isinstance(system.ruby._cpu_ports[i], VIPERCoalescer):
system.cpu[shader_idx].CUs[token_port_idx].gmTokenPort = \
system.ruby._cpu_ports[i].gmTokenPort
token_port_idx += 1
wavefront_size = args.wf_size
for i in range(args.num_compute_units):
# The pipeline issues wavefront_size number of uncoalesced requests
# in one GPU issue cycle. Hence wavefront_size mem ports.
for j in range(wavefront_size):
system.cpu[shader_idx].CUs[i].memory_port[j] = \
system.ruby._cpu_ports[gpu_port_idx].in_ports[j]
gpu_port_idx += 1
for i in range(args.num_compute_units):
if i > 0 and not i % args.cu_per_sqc:
gpu_port_idx += 1
system.cpu[shader_idx].CUs[i].sqc_port = \
system.ruby._cpu_ports[gpu_port_idx].in_ports
gpu_port_idx = gpu_port_idx + 1
for i in range(args.num_compute_units):
if i > 0 and not i % args.cu_per_scalar_cache:
gpu_port_idx += 1
system.cpu[shader_idx].CUs[i].scalar_port = \
system.ruby._cpu_ports[gpu_port_idx].in_ports
gpu_port_idx = gpu_port_idx + 1
return system
def build_test_system(np, simplessd):
cmdline = cmd_line_template()
if buildEnv['TARGET_ISA'] == "alpha":
test_sys = makeLinuxAlphaSystem(test_mem_mode,
bm[0],
options.ruby,
cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "mips":
test_sys = makeLinuxMipsSystem(test_mem_mode, bm[0], cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "sparc":
test_sys = makeSparcSystem(test_mem_mode, bm[0], cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "x86":
test_sys = makeLinuxX86System(test_mem_mode,
simplessd,
options.num_cpus,
bm[0],
options.ruby,
cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "arm":
test_sys = makeArmSystem(
test_mem_mode,
options.machine_type,
simplessd,
options.num_cpus,
bm[0],
options.dtb_filename,
bare_metal=options.bare_metal,
cmdline=cmdline,
external_memory=options.external_memory_system,
ruby=options.ruby,
security=options.enable_security_extensions)
if options.enable_context_switch_stats_dump:
test_sys.enable_context_switch_stats_dump = True
else:
fatal("Incapable of building %s full system!", buildEnv['TARGET_ISA'])
# Set the cache line size for the entire system
test_sys.cache_line_size = options.cacheline_size
# Create a top-level voltage domain
test_sys.voltage_domain = VoltageDomain(voltage=options.sys_voltage)
# Create a source clock for the system and set the clock period
test_sys.clk_domain = SrcClockDomain(
clock=options.sys_clock, voltage_domain=test_sys.voltage_domain)
# Create a CPU voltage domain
test_sys.cpu_voltage_domain = VoltageDomain()
# Create a source clock for the CPUs and set the clock period
test_sys.cpu_clk_domain = SrcClockDomain(
clock=options.cpu_clock, voltage_domain=test_sys.cpu_voltage_domain)
if options.kernel is not None:
test_sys.kernel = binary(options.kernel)
if options.script is not None:
test_sys.readfile = options.script
if options.lpae:
test_sys.have_lpae = True
if options.virtualisation:
test_sys.have_virtualization = True
test_sys.init_param = options.init_param
# For now, assign all the CPUs to the same clock domain
test_sys.cpu = [
TestCPUClass(clk_domain=test_sys.cpu_clk_domain, cpu_id=i)
for i in range(np)
]
if CpuConfig.is_kvm_cpu(TestCPUClass) or CpuConfig.is_kvm_cpu(FutureClass):
test_sys.kvm_vm = KvmVM()
if buildEnv['TARGET_ISA'] in "x86":
test_sys.eventq_index = np
if options.ruby:
bootmem = getattr(test_sys, 'bootmem', None)
Ruby.create_system(options, True, test_sys, test_sys.iobus,
test_sys._dma_ports, bootmem)
# Create a seperate clock domain for Ruby
test_sys.ruby.clk_domain = SrcClockDomain(
clock=options.ruby_clock, voltage_domain=test_sys.voltage_domain)
# Connect the ruby io port to the PIO bus,
# assuming that there is just one such port.
test_sys.iobus.master = test_sys.ruby._io_port.slave
for (i, cpu) in enumerate(test_sys.cpu):
#
# Tie the cpu ports to the correct ruby system ports
#
cpu.clk_domain = test_sys.cpu_clk_domain
cpu.createThreads()
cpu.createInterruptController()
cpu.icache_port = test_sys.ruby._cpu_ports[i].slave
cpu.dcache_port = test_sys.ruby._cpu_ports[i].slave
if buildEnv['TARGET_ISA'] in ("x86", "arm"):
cpu.itb.walker.port = test_sys.ruby._cpu_ports[i].slave
cpu.dtb.walker.port = test_sys.ruby._cpu_ports[i].slave
if buildEnv['TARGET_ISA'] in "x86":
cpu.interrupts[0].pio = test_sys.ruby._cpu_ports[i].master
cpu.interrupts[0].int_master = test_sys.ruby._cpu_ports[
i].slave
cpu.interrupts[0].int_slave = test_sys.ruby._cpu_ports[
i].master
else:
gicv2m_range = AddrRange(0x2c1c0000, 0x2c1d0000 - 1)
if options.caches or options.l2cache:
# By default the IOCache runs at the system clock
test_sys.iocache = IOCache(addr_ranges=test_sys.mem_ranges)
test_sys.iocache.cpu_side = test_sys.iobus.master
test_sys.iocache.mem_side = test_sys.membus.slave
if buildEnv['TARGET_ISA'] in "arm":
if options.machine_type == "VExpress_GEM5_V1":
test_sys.iobridge = Bridge(delay='50ns',
ranges=[gicv2m_range])
test_sys.iobridge.slave = test_sys.iobus.master
test_sys.iobridge.master = test_sys.membus.slave
elif not options.external_memory_system:
mem_ranges = list(test_sys.mem_ranges) # Copy list not reference
# Bypass MSI/MSI-X
if buildEnv['TARGET_ISA'] in "arm":
if options.machine_type == "VExpress_GEM5_V1":
mem_ranges.append(gicv2m_range)
test_sys.iobridge = Bridge(delay='50ns', ranges=mem_ranges)
test_sys.iobridge.slave = test_sys.iobus.master
test_sys.iobridge.master = test_sys.membus.slave
# Sanity check
if options.simpoint_profile:
if not CpuConfig.is_noncaching_cpu(TestCPUClass):
fatal("SimPoint generation should be done with atomic cpu")
if np > 1:
fatal(
"SimPoint generation not supported with more than one CPUs"
)
for i in range(np):
if options.simpoint_profile:
test_sys.cpu[i].addSimPointProbe(options.simpoint_interval)
if options.checker:
test_sys.cpu[i].addCheckerCpu()
if options.bp_type:
bpClass = BPConfig.get(options.bp_type)
test_sys.cpu[i].branchPred = bpClass()
test_sys.cpu[i].createThreads()
# If elastic tracing is enabled when not restoring from checkpoint and
# when not fast forwarding using the atomic cpu, then check that the
# TestCPUClass is DerivO3CPU or inherits from DerivO3CPU. If the check
# passes then attach the elastic trace probe.
# If restoring from checkpoint or fast forwarding, the code that does this for
# FutureCPUClass is in the Simulation module. If the check passes then the
# elastic trace probe is attached to the switch CPUs.
if options.elastic_trace_en and options.checkpoint_restore == None and \
not options.fast_forward:
CpuConfig.config_etrace(TestCPUClass, test_sys.cpu, options)
CacheConfig.config_cache(options, test_sys)
MemConfig.config_mem(options, test_sys)
if buildEnv['TARGET_ISA'] in "x86":
lapics = []
for i in range(np):
lapics.append(test_sys.cpu[i].interrupts[0])
test_sys.msi_handler.lapics = lapics
if buildEnv['TARGET_ISA'] in "x86" and test_sys.eventq_index == np:
test_sys.eventq_index = 0
for idx, cpu in enumerate(test_sys.cpu):
for obj in cpu.descendants():
obj.eventq_index = test_sys.eventq_index
cpu.eventq_index = idx + 1
return test_sys
for _, dma_device in enumerate(dma_devices):
dma_device.pio = system.piobus.mem_side_ports
system.dma_devices = dma_devices
#
# Create the Ruby system
#
# the ruby tester reuses num_cpus to specify the
# number of cpu ports connected to the tester object, which
# is stored in system.cpu. because there is only ever one
# tester object, num_cpus is not necessarily equal to the
# size of system.cpu
cpu_list = [system.cpu] * args.num_cpus
Ruby.create_system(args,
full_system=False,
system=system,
dma_ports=system.dma_devices,
cpus=cpu_list)
#
# The tester is most effective when randomization is turned on and
# artifical delay is randomly inserted on messages
#
system.ruby.randomization = True
# Assert that we got the right number of Ruby ports
assert (len(system.ruby._cpu_ports) == n_CUs + n_SQCs + n_Scalars)
#
# Attach Ruby ports to the tester in the order:
# cpu_sequencers,
system = System(cpu=tester)
# Dummy voltage domain for all our clock domains
system.voltage_domain = VoltageDomain(voltage=args.sys_voltage)
system.clk_domain = SrcClockDomain(clock='1GHz',
voltage_domain=system.voltage_domain)
system.mem_ranges = AddrRange('256MB')
# the ruby tester reuses num_cpus to specify the
# number of cpu ports connected to the tester object, which
# is stored in system.cpu. because there is only ever one
# tester object, num_cpus is not necessarily equal to the
# size of system.cpu
cpu_list = [system.cpu] * args.num_cpus
Ruby.create_system(args, False, system, cpus=cpu_list)
# Create a separate clock domain for Ruby
system.ruby.clk_domain = SrcClockDomain(clock='1GHz',
voltage_domain=system.voltage_domain)
tester.num_cpus = len(system.ruby._cpu_ports)
#
# The tester is most effective when randomization is turned on and
# artifical delay is randomly inserted on messages
#
system.ruby.randomization = True
for ruby_port in system.ruby._cpu_ports:
#
if args.num_dmas > 0:
dmas = [ MemTest(max_loads = args.maxloads,
percent_functional = 0,
percent_uncacheable = 0,
progress_interval = args.progress,
suppress_func_errors =
not args.suppress_func_errors) \
for i in range(args.num_dmas) ]
system.dma_devices = dmas
else:
dmas = []
dma_ports = []
for (i, dma) in enumerate(dmas):
dma_ports.append(dma.test)
Ruby.create_system(args, False, system, dma_ports = dma_ports)
# Create a top-level voltage domain and clock domain
system.voltage_domain = VoltageDomain(voltage = args.sys_voltage)
system.clk_domain = SrcClockDomain(clock = args.sys_clock,
voltage_domain = system.voltage_domain)
# Create a seperate clock domain for Ruby
system.ruby.clk_domain = SrcClockDomain(clock = args.ruby_clock,
voltage_domain = system.voltage_domain)
#
# The tester is most effective when randomization is turned on and
# artifical delay is randomly inserted on messages
#
system.ruby.randomization = True
def build_test_system(np):
cmdline = cmd_line_template()
if buildEnv['TARGET_ISA'] == "alpha":
test_sys = makeLinuxAlphaSystem(test_mem_mode,
bm[0],
options.ruby,
cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "mips":
test_sys = makeLinuxMipsSystem(test_mem_mode, bm[0], cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "sparc":
test_sys = makeSparcSystem(test_mem_mode, bm[0], cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "x86":
test_sys = makeLinuxX86System(test_mem_mode,
options.num_cpus,
bm[0],
options.ruby,
cmdline=cmdline,
cossim_enabled=options.cossim,
nodeNum=options.nodeNum)
elif buildEnv['TARGET_ISA'] == "arm":
test_sys = makeArmSystem(
test_mem_mode,
options.machine_type,
options.num_cpus,
bm[0],
options.dtb_filename,
bare_metal=options.bare_metal,
cmdline=cmdline,
external_memory=options.external_memory_system,
ruby=options.ruby,
security=options.enable_security_extensions,
cossim_enabled=options.cossim,
nodeNum=options.nodeNum,
system_clock=options.sys_clock,
dev_clk=options.device_clock)
if options.enable_context_switch_stats_dump:
test_sys.enable_context_switch_stats_dump = True
else:
fatal("Incapable of building %s full system!", buildEnv['TARGET_ISA'])
# Set the cache line size for the entire system
test_sys.cache_line_size = options.cacheline_size
# Create a top-level voltage domain
test_sys.voltage_domain = VoltageDomain(voltage=options.sys_voltage)
# Create a source clock for the system and set the clock period
test_sys.clk_domain = SrcClockDomain(
clock=options.sys_clock, voltage_domain=test_sys.voltage_domain)
# Create a CPU voltage domain
test_sys.cpu_voltage_domain = VoltageDomain()
# Create a source clock for the CPUs and set the clock period
test_sys.cpu_clk_domain = SrcClockDomain(
clock=options.cpu_clock, voltage_domain=test_sys.cpu_voltage_domain)
if options.kernel is not None:
test_sys.kernel = binary(options.kernel)
if options.script is not None:
test_sys.readfile = options.script
if options.lpae:
test_sys.have_lpae = True
if options.virtualisation:
test_sys.have_virtualization = True
test_sys.init_param = options.init_param
# For now, assign all the CPUs to the same clock domain
test_sys.cpu = [
TestCPUClass(clk_domain=test_sys.cpu_clk_domain, cpu_id=i)
for i in xrange(np)
]
if is_kvm_cpu(TestCPUClass) or is_kvm_cpu(FutureClass):
test_sys.kvm_vm = KvmVM()
if options.ruby:
# Check for timing mode because ruby does not support atomic accesses
if not (options.cpu_type == "DerivO3CPU"
or options.cpu_type == "TimingSimpleCPU"):
print >> sys.stderr, "Ruby requires TimingSimpleCPU or O3CPU!!"
sys.exit(1)
Ruby.create_system(options, True, test_sys, test_sys.iobus,
test_sys._dma_ports)
# Create a seperate clock domain for Ruby
test_sys.ruby.clk_domain = SrcClockDomain(
clock=options.ruby_clock, voltage_domain=test_sys.voltage_domain)
# Connect the ruby io port to the PIO bus,
# assuming that there is just one such port.
test_sys.iobus.master = test_sys.ruby._io_port.slave
for (i, cpu) in enumerate(test_sys.cpu):
#
# Tie the cpu ports to the correct ruby system ports
#
cpu.clk_domain = test_sys.cpu_clk_domain
cpu.createThreads()
cpu.createInterruptController()
cpu.icache_port = test_sys.ruby._cpu_ports[i].slave
cpu.dcache_port = test_sys.ruby._cpu_ports[i].slave
if buildEnv['TARGET_ISA'] in ("x86", "arm"):
cpu.itb.walker.port = test_sys.ruby._cpu_ports[i].slave
cpu.dtb.walker.port = test_sys.ruby._cpu_ports[i].slave
if buildEnv['TARGET_ISA'] in "x86":
cpu.interrupts[0].pio = test_sys.ruby._cpu_ports[i].master
cpu.interrupts[0].int_master = test_sys.ruby._cpu_ports[
i].slave
cpu.interrupts[0].int_slave = test_sys.ruby._cpu_ports[
i].master
else:
if options.caches or options.l2cache:
# By default the IOCache runs at the system clock
test_sys.iocache = IOCache(addr_ranges=test_sys.mem_ranges)
test_sys.iocache.cpu_side = test_sys.iobus.master
test_sys.iocache.mem_side = test_sys.membus.slave
elif not options.external_memory_system:
test_sys.iobridge = Bridge(delay='50ns',
ranges=test_sys.mem_ranges)
test_sys.iobridge.slave = test_sys.iobus.master
test_sys.iobridge.master = test_sys.membus.slave
# Sanity check
if options.fastmem:
if TestCPUClass != AtomicSimpleCPU:
fatal("Fastmem can only be used with atomic CPU!")
if (options.caches or options.l2cache):
fatal("You cannot use fastmem in combination with caches!")
if options.simpoint_profile:
if not options.fastmem:
# Atomic CPU checked with fastmem option already
fatal(
"SimPoint generation should be done with atomic cpu and fastmem"
)
if np > 1:
fatal(
"SimPoint generation not supported with more than one CPUs"
)
for i in xrange(np):
if options.fastmem:
test_sys.cpu[i].fastmem = True
if options.simpoint_profile:
test_sys.cpu[i].addSimPointProbe(options.simpoint_interval)
if options.checker:
test_sys.cpu[i].addCheckerCpu()
test_sys.cpu[i].createThreads()
# If elastic tracing is enabled when not restoring from checkpoint and
# when not fast forwarding using the atomic cpu, then check that the
# TestCPUClass is DerivO3CPU or inherits from DerivO3CPU. If the check
# passes then attach the elastic trace probe.
# If restoring from checkpoint or fast forwarding, the code that does this for
# FutureCPUClass is in the Simulation module. If the check passes then the
# elastic trace probe is attached to the switch CPUs.
if options.elastic_trace_en and options.checkpoint_restore == None and \
not options.fast_forward:
CpuConfig.config_etrace(TestCPUClass, test_sys.cpu, options)
CacheConfig.config_cache(options, test_sys)
MemConfig.config_mem(options, test_sys)
return test_sys
mdesc = SysConfig(disks = ['linux-x86.img'])
system = FSConfig.makeLinuxX86System('timing', args.num_cpus,
mdesc=mdesc, Ruby=True)
system.kernel = SysPaths.binary('x86_64-vmlinux-2.6.22.9')
# Dummy voltage domain for all our clock domains
system.voltage_domain = VoltageDomain(voltage = args.sys_voltage)
system.kernel = FSConfig.binary('x86_64-vmlinux-2.6.22.9.smp')
system.clk_domain = SrcClockDomain(clock = '1GHz',
voltage_domain = system.voltage_domain)
system.cpu_clk_domain = SrcClockDomain(clock = '2GHz',
voltage_domain = system.voltage_domain)
system.cpu = [TimingSimpleCPU(cpu_id=i, clk_domain = system.cpu_clk_domain)
for i in range(args.num_cpus)]
Ruby.create_system(args, True, system, system.iobus, system._dma_ports)
# Create a seperate clock domain for Ruby
system.ruby.clk_domain = SrcClockDomain(clock = args.ruby_clock,
voltage_domain = system.voltage_domain)
# Connect the ruby io port to the PIO bus,
# assuming that there is just one such port.
system.iobus.mem_side_ports = system.ruby._io_port.in_ports
for (i, cpu) in enumerate(system.cpu):
# create the interrupt controller
cpu.createInterruptController()
# Tie the cpu ports to the correct ruby system ports
system.ruby._cpu_ports[i].connectCpuPorts(cpu)
if options.fastmem:
system.cpu[i].fastmem = True
if options.simpoint_profile:
system.cpu[i].simpoint_profile = True
system.cpu[i].addSimPointProbe(options.simpoint_interval)
if options.checker:
system.cpu[i].addCheckerCpu()
system.cpu[i].createThreads()
if options.ruby:
#print(">> Not here")
options.use_map = True
Ruby.create_system(options, len(multiprocesses), False, system)
assert(options.num_cpus == len(system.ruby._cpu_ports))
system.ruby.clk_domain = SrcClockDomain(clock = options.ruby_clock,
voltage_domain = system.voltage_domain)
for i in xrange(np):
ruby_port = system.ruby._cpu_ports[i]
# Create the interrupt controller and connect its ports to Ruby
# Note that the interrupt controller is always present but only
# in x86 does it have message ports that need to be connected
system.cpu[i].createInterruptController()
# Connect the cpu's cache ports to Ruby
system.cpu[i].icache_port = ruby_port.slave
system.cpu[i].dcache_port = ruby_port.slave
def build_test_system(np):
cmdline = cmd_line_template()
if buildEnv['TARGET_ISA'] == "mips":
test_sys = makeLinuxMipsSystem(test_mem_mode, bm[0], cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "sparc":
test_sys = makeSparcSystem(test_mem_mode, bm[0], cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "riscv":
test_sys = makeBareMetalRiscvSystem(test_mem_mode,
bm[0],
cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "x86":
test_sys = makeLinuxX86System(test_mem_mode,
np,
bm[0],
args.ruby,
cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "arm":
test_sys = makeArmSystem(
test_mem_mode,
args.machine_type,
np,
bm[0],
args.dtb_filename,
bare_metal=args.bare_metal,
cmdline=cmdline,
external_memory=args.external_memory_system,
ruby=args.ruby,
vio_9p=args.vio_9p,
bootloader=args.bootloader,
)
if args.enable_context_switch_stats_dump:
test_sys.enable_context_switch_stats_dump = True
else:
fatal("Incapable of building %s full system!", buildEnv['TARGET_ISA'])
# Set the cache line size for the entire system
test_sys.cache_line_size = args.cacheline_size
# Create a top-level voltage domain
test_sys.voltage_domain = VoltageDomain(voltage=args.sys_voltage)
# Create a source clock for the system and set the clock period
test_sys.clk_domain = SrcClockDomain(
clock=args.sys_clock, voltage_domain=test_sys.voltage_domain)
# Create a CPU voltage domain
test_sys.cpu_voltage_domain = VoltageDomain()
# Create a source clock for the CPUs and set the clock period
test_sys.cpu_clk_domain = SrcClockDomain(
clock=args.cpu_clock, voltage_domain=test_sys.cpu_voltage_domain)
if buildEnv['TARGET_ISA'] == 'riscv':
test_sys.workload.bootloader = args.kernel
elif args.kernel is not None:
test_sys.workload.object_file = binary(args.kernel)
if args.script is not None:
test_sys.readfile = args.script
test_sys.init_param = args.init_param
# For now, assign all the CPUs to the same clock domain
test_sys.cpu = [
TestCPUClass(clk_domain=test_sys.cpu_clk_domain, cpu_id=i)
for i in range(np)
]
if args.ruby:
bootmem = getattr(test_sys, '_bootmem', None)
Ruby.create_system(args, True, test_sys, test_sys.iobus,
test_sys._dma_ports, bootmem)
# Create a seperate clock domain for Ruby
test_sys.ruby.clk_domain = SrcClockDomain(
clock=args.ruby_clock, voltage_domain=test_sys.voltage_domain)
# Connect the ruby io port to the PIO bus,
# assuming that there is just one such port.
test_sys.iobus.mem_side_ports = test_sys.ruby._io_port.in_ports
for (i, cpu) in enumerate(test_sys.cpu):
#
# Tie the cpu ports to the correct ruby system ports
#
cpu.clk_domain = test_sys.cpu_clk_domain
cpu.createThreads()
cpu.createInterruptController()
test_sys.ruby._cpu_ports[i].connectCpuPorts(cpu)
else:
if args.caches or args.l2cache:
# By default the IOCache runs at the system clock
test_sys.iocache = IOCache(addr_ranges=test_sys.mem_ranges)
test_sys.iocache.cpu_side = test_sys.iobus.mem_side_ports
test_sys.iocache.mem_side = test_sys.membus.cpu_side_ports
elif not args.external_memory_system:
test_sys.iobridge = Bridge(delay='50ns',
ranges=test_sys.mem_ranges)
test_sys.iobridge.cpu_side_port = test_sys.iobus.mem_side_ports
test_sys.iobridge.mem_side_port = test_sys.membus.cpu_side_ports
# Sanity check
if args.simpoint_profile:
if not ObjectList.is_noncaching_cpu(TestCPUClass):
fatal("SimPoint generation should be done with atomic cpu")
if np > 1:
fatal(
"SimPoint generation not supported with more than one CPUs"
)
for i in range(np):
if args.simpoint_profile:
test_sys.cpu[i].addSimPointProbe(args.simpoint_interval)
if args.checker:
test_sys.cpu[i].addCheckerCpu()
if not ObjectList.is_kvm_cpu(TestCPUClass):
if args.bp_type:
bpClass = ObjectList.bp_list.get(args.bp_type)
test_sys.cpu[i].branchPred = bpClass()
if args.indirect_bp_type:
IndirectBPClass = ObjectList.indirect_bp_list.get(
args.indirect_bp_type)
test_sys.cpu[i].branchPred.indirectBranchPred = \
IndirectBPClass()
test_sys.cpu[i].createThreads()
# If elastic tracing is enabled when not restoring from checkpoint and
# when not fast forwarding using the atomic cpu, then check that the
# TestCPUClass is DerivO3CPU or inherits from DerivO3CPU. If the check
# passes then attach the elastic trace probe.
# If restoring from checkpoint or fast forwarding, the code that does this for
# FutureCPUClass is in the Simulation module. If the check passes then the
# elastic trace probe is attached to the switch CPUs.
if args.elastic_trace_en and args.checkpoint_restore == None and \
not args.fast_forward:
CpuConfig.config_etrace(TestCPUClass, test_sys.cpu, args)
CacheConfig.config_cache(args, test_sys)
MemConfig.config_mem(args, test_sys)
if ObjectList.is_kvm_cpu(TestCPUClass) or \
ObjectList.is_kvm_cpu(FutureClass):
# Assign KVM CPUs to their own event queues / threads. This
# has to be done after creating caches and other child objects
# since these mustn't inherit the CPU event queue.
for i, cpu in enumerate(test_sys.cpu):
# Child objects usually inherit the parent's event
# queue. Override that and use the same event queue for
# all devices.
for obj in cpu.descendants():
obj.eventq_index = 0
cpu.eventq_index = i + 1
test_sys.kvm_vm = KvmVM()
return test_sys
system.clk_domain = SrcClockDomain(clock = '1GHz',
voltage_domain = system.voltage_domain)
# Create a seperate clock domain for components that should run at
# CPUs frequency
system.cpu[0].clk_domain = SrcClockDomain(clock = '2GHz',
voltage_domain = \
system.voltage_domain)
# configure the TLB hierarchy
GPUTLBConfig.config_tlb_hierarchy(options, system, shader_idx)
# create Ruby system
system.piobus = IOXBar(width=32, response_latency=0,
frontend_latency=0, forward_latency=0)
Ruby.create_system(options, None, system)
# Create a separate clock for Ruby
system.ruby.clk_domain = SrcClockDomain(clock = options.ruby_clock,
voltage_domain = system.voltage_domain)
# create the interrupt controller
cpu.createInterruptController()
#
# Tie the cpu cache ports to the ruby cpu ports and
# physmem, respectively
#
cpu.connectAllPorts(system.ruby._cpu_ports[0])
system.ruby._cpu_ports[0].mem_master_port = system.piobus.slave
def build_test_system(np):
cmdline = cmd_line_template()
if buildEnv['TARGET_ISA'] == "alpha":
test_sys = makeLinuxAlphaSystem(test_mem_mode,
bm[0],
options.ruby,
cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "mips":
test_sys = makeLinuxMipsSystem(test_mem_mode, bm[0], cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "sparc":
test_sys = makeSparcSystem(test_mem_mode, bm[0], cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "x86":
test_sys = makeLinuxX86System(test_mem_mode,
options.num_cpus,
bm[0],
options.ruby,
cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "arm":
test_sys = makeArmSystem(
test_mem_mode,
options.machine_type,
options.membus_width,
options,
options.num_cpus,
bm[0],
options.dtb_filename,
bare_metal=options.bare_metal,
cmdline=cmdline,
ignore_dtb=options.generate_dtb,
external_memory=options.external_memory_system,
ruby=options.ruby,
security=options.enable_security_extensions)
if options.enable_context_switch_stats_dump:
test_sys.enable_context_switch_stats_dump = True
else:
fatal("Incapable of building %s full system!", buildEnv['TARGET_ISA'])
# Set the cache line size for the entire system
test_sys.cache_line_size = options.cacheline_size
# Create a top-level voltage domain
test_sys.voltage_domain = VoltageDomain(voltage=options.sys_voltage)
# Create a source clock for the system and set the clock period
test_sys.clk_domain = SrcClockDomain(
clock=options.sys_clock, voltage_domain=test_sys.voltage_domain)
# Create a CPU voltage domain
test_sys.cpu_voltage_domain = VoltageDomain()
# Create a source clock for the CPUs and set the clock period
test_sys.cpu_clk_domain = SrcClockDomain(
clock=options.cpu_clock, voltage_domain=test_sys.cpu_voltage_domain)
if options.kernel is not None:
test_sys.kernel = binary(options.kernel)
if options.script is not None:
test_sys.readfile = options.script
if options.lpae:
test_sys.have_lpae = True
if options.virtualisation:
test_sys.have_virtualization = True
test_sys.init_param = options.init_param
# For now, assign all the CPUs to the same clock domain
test_sys.cpu = [
TestCPUClass(clk_domain=test_sys.cpu_clk_domain, cpu_id=i)
for i in xrange(np)
]
if is_kvm_cpu(TestCPUClass) or is_kvm_cpu(FutureClass):
test_sys.kvm_vm = KvmVM()
if options.ruby:
bootmem = getattr(test_sys, 'bootmem', None)
Ruby.create_system(options, True, test_sys, test_sys.iobus,
test_sys._dma_ports, bootmem)
# Create a seperate clock domain for Ruby
test_sys.ruby.clk_domain = SrcClockDomain(
clock=options.ruby_clock, voltage_domain=test_sys.voltage_domain)
# Connect the ruby io port to the PIO bus,
# assuming that there is just one such port.
test_sys.iobus.master = test_sys.ruby._io_port.slave
for (i, cpu) in enumerate(test_sys.cpu):
#
# Tie the cpu ports to the correct ruby system ports
#
cpu.clk_domain = test_sys.cpu_clk_domain
cpu.createThreads()
cpu.createInterruptController()
cpu.icache_port = test_sys.ruby._cpu_ports[i].slave
cpu.dcache_port = test_sys.ruby._cpu_ports[i].slave
if buildEnv['TARGET_ISA'] in ("x86", "arm"):
cpu.itb.walker.port = test_sys.ruby._cpu_ports[i].slave
cpu.dtb.walker.port = test_sys.ruby._cpu_ports[i].slave
if buildEnv['TARGET_ISA'] in "x86":
cpu.interrupts[0].pio = test_sys.ruby._cpu_ports[i].master
cpu.interrupts[0].int_master = test_sys.ruby._cpu_ports[
i].slave
cpu.interrupts[0].int_slave = test_sys.ruby._cpu_ports[
i].master
else:
if options.caches or options.l2cache:
# By default the IOCache runs at the system clock
test_sys.iocache = IOCache(addr_ranges=test_sys.mem_ranges)
test_sys.iocache.cpu_side = test_sys.iobus.master
test_sys.iocache.mem_side = test_sys.membus.slave
elif not options.external_memory_system:
test_sys.iobridge = Bridge(delay='50ns',
ranges=test_sys.mem_ranges)
test_sys.iobridge.slave = test_sys.iobus.master
test_sys.iobridge.master = test_sys.membus.slave
# Sanity check
if options.fastmem:
if TestCPUClass != AtomicSimpleCPU:
fatal("Fastmem can only be used with atomic CPU!")
if (options.caches or options.l2cache):
fatal("You cannot use fastmem in combination with caches!")
if options.simpoint_profile:
if not options.fastmem:
# Atomic CPU checked with fastmem option already
fatal(
"SimPoint generation should be done with atomic cpu and fastmem"
)
if np > 1:
fatal(
"SimPoint generation not supported with more than one CPUs"
)
for i in xrange(np):
if options.fastmem:
test_sys.cpu[i].fastmem = True
if options.simpoint_profile:
test_sys.cpu[i].addSimPointProbe(options.simpoint_interval)
if options.checker:
test_sys.cpu[i].addCheckerCpu()
test_sys.cpu[i].createThreads()
# If elastic tracing is enabled when not restoring from checkpoint and
# when not fast forwarding using the atomic cpu, then check that the
# TestCPUClass is DerivO3CPU or inherits from DerivO3CPU. If the check
# passes then attach the elastic trace probe.
# If restoring from checkpoint or fast forwarding, the code that does this for
# FutureCPUClass is in the Simulation module. If the check passes then the
# elastic trace probe is attached to the switch CPUs.
if options.elastic_trace_en and options.checkpoint_restore == None and \
not options.fast_forward:
CpuConfig.config_etrace(TestCPUClass, test_sys.cpu, options)
if buildEnv['TARGET_ISA'] != "arm" and options.workload_automation_vio:
warn("Ignoring --workload-automation-vio. It is unsupported on "
"non-ARM systems.")
elif options.workload_automation_vio:
from m5.objects import PciVirtIO, VirtIO9PDiod
viopci = PciVirtIO(
pci_bus=0,
pci_dev=test_sys.realview._num_pci_dev,
pci_func=0,
InterruptPin=1,
InterruptLine=test_sys.realview._num_pci_int_line)
test_sys.realview._num_pci_dev = test_sys.realview._num_pci_dev + 1
test_sys.realview._num_pci_int_line = test_sys.realview._num_pci_int_line + 1
viopci.vio = VirtIO9PDiod()
viopci.vio.root = options.workload_automation_vio
viopci.vio.socketPath = "/home/yqureshi/shares/local/scrap/temp"
test_sys.realview.viopci = viopci
test_sys.realview.viopci.dma = test_sys.iobus.slave
test_sys.realview.viopci.pio = test_sys.iobus.master
CacheConfig.config_cache(options, test_sys)
MemConfig.config_mem(options, test_sys)
return test_sys
#the system
mdesc = SysConfig(disk = 'linux-x86.img')
system = FSConfig.makeLinuxX86System('timing', options.num_cpus,
mdesc=mdesc, Ruby=True)
# Dummy voltage domain for all our clock domains
system.voltage_domain = VoltageDomain(voltage = options.sys_voltage)
system.kernel = FSConfig.binary('x86_64-vmlinux-2.6.22.9.smp')
system.clk_domain = SrcClockDomain(clock = '1GHz',
voltage_domain = system.voltage_domain)
system.cpu_clk_domain = SrcClockDomain(clock = '2GHz',
voltage_domain = system.voltage_domain)
system.cpu = [TimingSimpleCPU(cpu_id=i, clk_domain = system.cpu_clk_domain)
for i in xrange(options.num_cpus)]
Ruby.create_system(options, True, system, system.iobus, system._dma_ports)
# Create a seperate clock domain for Ruby
system.ruby.clk_domain = SrcClockDomain(clock = options.ruby_clock,
voltage_domain = system.voltage_domain)
# Connect the ruby io port to the PIO bus,
# assuming that there is just one such port.
system.iobus.master = system.ruby._io_port.slave
for (i, cpu) in enumerate(system.cpu):
# create the interrupt controller
cpu.createInterruptController()
# Tie the cpu ports to the correct ruby system ports
cpu.icache_port = system.ruby._cpu_ports[i].slave
cpu.dcache_port = system.ruby._cpu_ports[i].slave
max_loads = options.maxloads,
issue_dmas = True,
percent_functional = 0,
percent_uncacheable = 0,
progress_interval = options.progress,
suppress_func_warnings =
not options.suppress_func_warnings) \
for i in xrange(options.num_dmas) ]
system.dma_devices = dmas
else:
dmas = []
dma_ports = []
for (i, dma) in enumerate(dmas):
dma_ports.append(dma.test)
Ruby.create_system(options, False, system, dma_ports = dma_ports)
# Create a top-level voltage domain and clock domain
system.voltage_domain = VoltageDomain(voltage = options.sys_voltage)
system.clk_domain = SrcClockDomain(clock = options.sys_clock,
voltage_domain = system.voltage_domain)
# Create a seperate clock domain for Ruby
system.ruby.clk_domain = SrcClockDomain(clock = options.ruby_clock,
voltage_domain = system.voltage_domain)
#
# The tester is most effective when randomization is turned on and
# artifical delay is randomly inserted on messages
#
system.ruby.randomization = True
system.vm = KvmVM()
for i in xrange(len(host_cpu.workload)):
host_cpu.workload[i].useArchPT = True
host_cpu.workload[i].kvmInSE = True
else:
fatal("KvmCPU can only be used in SE mode with x86")
# configure the TLB hierarchy
GPUTLBConfig.config_tlb_hierarchy(options, system, shader_idx)
# create Ruby system
system.piobus = IOXBar(width=32,
response_latency=0,
frontend_latency=0,
forward_latency=0)
Ruby.create_system(options, None, system)
system.ruby.clk_domain = SrcClockDomain(clock=options.ruby_clock,
voltage_domain=system.voltage_domain)
# attach the CPU ports to Ruby
for i in range(options.num_cpus):
ruby_port = system.ruby._cpu_ports[i]
# Create interrupt controller
system.cpu[i].createInterruptController()
# Connect cache port's to ruby
system.cpu[i].icache_port = ruby_port.slave
system.cpu[i].dcache_port = ruby_port.slave
ruby_port.mem_master_port = system.piobus.slave
def build_test_system(np):
cmdline = cmd_line_template()
if buildEnv['TARGET_ISA'] == "mips":
test_sys = makeLinuxMipsSystem(test_mem_mode, bm[0], cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "sparc":
test_sys = makeSparcSystem(test_mem_mode, bm[0], cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "riscv":
test_sys = makeBareMetalRiscvSystem(test_mem_mode,
bm[0],
cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "x86":
test_sys = makeLinuxX86System(test_mem_mode,
np,
bm[0],
options.ruby,
cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "arm":
test_sys = makeArmSystem(
test_mem_mode,
options.machine_type,
np,
bm[0],
options.dtb_filename,
bare_metal=options.bare_metal,
cmdline=cmdline,
external_memory=options.external_memory_system,
ruby=options.ruby,
security=options.enable_security_extensions,
vio_9p=options.vio_9p,
bootloader=options.bootloader,
)
if options.enable_context_switch_stats_dump:
test_sys.enable_context_switch_stats_dump = True
else:
fatal("Incapable of building %s full system!", buildEnv['TARGET_ISA'])
# Set the cache line size for the entire system
test_sys.cache_line_size = options.cacheline_size
# Create a top-level voltage domain
test_sys.voltage_domain = VoltageDomain(voltage=options.sys_voltage)
# Create a source clock for the system and set the clock period
test_sys.clk_domain = SrcClockDomain(
clock=options.sys_clock, voltage_domain=test_sys.voltage_domain)
# Create a CPU voltage domain
test_sys.cpu_voltage_domain = VoltageDomain()
# Create a source clock for the CPUs and set the clock period
test_sys.cpu_clk_domain = SrcClockDomain(
clock=options.cpu_clock, voltage_domain=test_sys.cpu_voltage_domain)
if buildEnv['TARGET_ISA'] == 'riscv':
test_sys.workload.bootloader = options.kernel
elif options.kernel is not None:
test_sys.workload.object_file = binary(options.kernel)
if options.script is not None:
test_sys.readfile = options.script
if options.lpae:
test_sys.have_lpae = True
if options.virtualisation:
test_sys.have_virtualization = True
test_sys.init_param = options.init_param
# For now, assign all the CPUs to the same clock domain
test_sys.cpu = [
TestCPUClass(clk_domain=test_sys.cpu_clk_domain, cpu_id=i)
for i in range(np)
]
if ObjectList.is_kvm_cpu(TestCPUClass) or \
ObjectList.is_kvm_cpu(FutureClass):
test_sys.kvm_vm = KvmVM()
if options.ruby:
bootmem = getattr(test_sys, '_bootmem', None)
Ruby.create_system(options, True, test_sys, test_sys.iobus,
test_sys._dma_ports, bootmem)
# Create a seperate clock domain for Ruby
test_sys.ruby.clk_domain = SrcClockDomain(
clock=options.ruby_clock, voltage_domain=test_sys.voltage_domain)
# Connect the ruby io port to the PIO bus,
# assuming that there is just one such port.
test_sys.iobus.master = test_sys.ruby._io_port.slave
for (i, cpu) in enumerate(test_sys.cpu):
#
# Tie the cpu ports to the correct ruby system ports
#
cpu.clk_domain = test_sys.cpu_clk_domain
cpu.createThreads()
cpu.createInterruptController()
cpu.icache_port = test_sys.ruby._cpu_ports[i].slave
cpu.dcache_port = test_sys.ruby._cpu_ports[i].slave
if buildEnv['TARGET_ISA'] in ("x86", "arm"):
cpu.itb.walker.port = test_sys.ruby._cpu_ports[i].slave
cpu.dtb.walker.port = test_sys.ruby._cpu_ports[i].slave
if buildEnv['TARGET_ISA'] in "x86":
cpu.interrupts[0].pio = test_sys.ruby._cpu_ports[i].master
cpu.interrupts[0].int_master = test_sys.ruby._cpu_ports[
i].slave
cpu.interrupts[0].int_slave = test_sys.ruby._cpu_ports[
i].master
else:
if options.caches or options.l2cache:
# By default the IOCache runs at the system clock
test_sys.iocache = IOCache(addr_ranges=test_sys.mem_ranges)
test_sys.iocache.cpu_side = test_sys.iobus.master
test_sys.iocache.mem_side = test_sys.membus.slave
elif not options.external_memory_system:
test_sys.iobridge = Bridge(delay='50ns',
ranges=test_sys.mem_ranges)
test_sys.iobridge.slave = test_sys.iobus.master
test_sys.iobridge.master = test_sys.membus.slave
# Sanity check
if options.simpoint_profile:
if not ObjectList.is_noncaching_cpu(TestCPUClass):
fatal("SimPoint generation should be done with atomic cpu")
if np > 1:
fatal(
"SimPoint generation not supported with more than one CPUs"
)
for i in range(np):
if options.simpoint_profile:
test_sys.cpu[i].addSimPointProbe(options.simpoint_interval)
if options.checker:
test_sys.cpu[i].addCheckerCpu()
if not ObjectList.is_kvm_cpu(TestCPUClass):
if options.bp_type:
bpClass = ObjectList.bp_list.get(options.bp_type)
test_sys.cpu[i].branchPred = bpClass()
if options.indirect_bp_type:
IndirectBPClass = ObjectList.indirect_bp_list.get(
options.indirect_bp_type)
test_sys.cpu[i].branchPred.indirectBranchPred = \
IndirectBPClass()
test_sys.cpu[i].createThreads()
# If elastic tracing is enabled when not restoring from checkpoint and
# when not fast forwarding using the atomic cpu, then check that the
# TestCPUClass is DerivO3CPU or inherits from DerivO3CPU. If the check
# passes then attach the elastic trace probe.
# If restoring from checkpoint or fast forwarding, the code that does this for
# FutureCPUClass is in the Simulation module. If the check passes then the
# elastic trace probe is attached to the switch CPUs.
if options.elastic_trace_en and options.checkpoint_restore == None and \
not options.fast_forward:
CpuConfig.config_etrace(TestCPUClass, test_sys.cpu, options)
CacheConfig.config_cache(options, test_sys)
MemConfig.config_mem(options, test_sys)
'''
pmu_listeners = []
for cpu in test_sys.cpu:
if buildEnv['TARGET_ISA'] in "arm":
for isa in cpu.isa:
isa.pmu = ArmPMU(interrupt=ArmPPI(num=20))
isa.pmu.addArchEvents(
cpu=cpu, dtb=cpu.dtb, itb=cpu.itb,
icache=getattr(cpu, "icache", None),
dcache=getattr(cpu, "dcache", None),
l2cache=getattr(test_sys, "l2", None))
pmu_listeners.append(PmuListener(
cpu=cpu,
dtb=cpu.dtb,
itb=cpu.itb,
icache=getattr(cpu, "icache", None),
dcache=getattr(cpu, "dcache", None),
l2cache=getattr(test_sys, "l2", None),
bpred=getattr(cpu, "branchPred", None),
))
test_sys.pmu_listeners = pmu_listeners
'''
for cpu in test_sys.cpu:
if buildEnv['TARGET_ISA'] in "arm":
for isa in cpu.isa:
isa.pmu = ArmPMU(interrupt=ArmPPI(num=20))
isa.pmu.addArchEvents(cpu=cpu,
dtb=cpu.dtb,
itb=cpu.itb,
icache=getattr(cpu, "icache", None),
dcache=getattr(cpu, "dcache", None),
l2cache=getattr(test_sys, "l2", None))
return test_sys
options.num_cp = 0
#
# Make generic DMA sequencer for Ruby to use
#
dma_devices = [TesterDma()] * n_DMAs
system.piobus = IOXBar()
for _, dma_device in enumerate(dma_devices):
dma_device.pio = system.piobus.mem_side_ports
system.dma_devices = dma_devices
#
# Create the Ruby system
#
Ruby.create_system(options=options,
full_system=False,
system=system,
dma_ports=system.dma_devices)
#
# The tester is most effective when randomization is turned on and
# artifical delay is randomly inserted on messages
#
system.ruby.randomization = True
# Assert that we got the right number of Ruby ports
assert (len(system.ruby._cpu_ports) == n_CUs + n_SQCs + n_Scalars)
#
# Attach Ruby ports to the tester in the order:
# cpu_sequencers,
# vector_coalescers,
def build_test_system(np):
cmdline = cmd_line_template()
if buildEnv['TARGET_ISA'] == "alpha":
test_sys = makeLinuxAlphaSystem(test_mem_mode, bm[0], options.ruby,
cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "mips":
test_sys = makeLinuxMipsSystem(test_mem_mode, bm[0], cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "sparc":
test_sys = makeSparcSystem(test_mem_mode, bm[0], cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "x86":
test_sys = makeLinuxX86System(test_mem_mode, np, bm[0], options.ruby,
cmdline=cmdline)
elif buildEnv['TARGET_ISA'] == "arm":
test_sys = makeArmSystem(test_mem_mode, options.machine_type, np,
bm[0], options.dtb_filename,
bare_metal=options.bare_metal,
cmdline=cmdline,
external_memory=
options.external_memory_system,
ruby=options.ruby,
security=options.enable_security_extensions)
if options.enable_context_switch_stats_dump:
test_sys.enable_context_switch_stats_dump = True
else:
fatal("Incapable of building %s full system!", buildEnv['TARGET_ISA'])
# Set the cache line size for the entire system
test_sys.cache_line_size = options.cacheline_size
# Create a top-level voltage domain
test_sys.voltage_domain = VoltageDomain(voltage = options.sys_voltage)
# Create a source clock for the system and set the clock period
test_sys.clk_domain = SrcClockDomain(clock = options.sys_clock,
voltage_domain = test_sys.voltage_domain)
# Create a CPU voltage domain
test_sys.cpu_voltage_domain = VoltageDomain()
# Create a source clock for the CPUs and set the clock period
test_sys.cpu_clk_domain = SrcClockDomain(clock = options.cpu_clock,
voltage_domain =
test_sys.cpu_voltage_domain)
if options.kernel is not None:
test_sys.kernel = binary(options.kernel)
else:
print("Error: a kernel must be provided to run in full system mode")
sys.exit(1)
if options.script is not None:
test_sys.readfile = options.script
if options.lpae:
test_sys.have_lpae = True
if options.virtualisation:
test_sys.have_virtualization = True
test_sys.init_param = options.init_param
# For now, assign all the CPUs to the same clock domain
test_sys.cpu = [TestCPUClass(clk_domain=test_sys.cpu_clk_domain, cpu_id=i)
for i in range(np)]
if CpuConfig.is_kvm_cpu(TestCPUClass) or CpuConfig.is_kvm_cpu(FutureClass):
test_sys.kvm_vm = KvmVM()
if options.ruby:
bootmem = getattr(test_sys, 'bootmem', None)
Ruby.create_system(options, True, test_sys, test_sys.iobus,
test_sys._dma_ports, bootmem)
# Create a seperate clock domain for Ruby
test_sys.ruby.clk_domain = SrcClockDomain(clock = options.ruby_clock,
voltage_domain = test_sys.voltage_domain)
# Connect the ruby io port to the PIO bus,
# assuming that there is just one such port.
test_sys.iobus.master = test_sys.ruby._io_port.slave
for (i, cpu) in enumerate(test_sys.cpu):
#
# Tie the cpu ports to the correct ruby system ports
#
cpu.clk_domain = test_sys.cpu_clk_domain
cpu.createThreads()
cpu.createInterruptController()
cpu.icache_port = test_sys.ruby._cpu_ports[i].slave
cpu.dcache_port = test_sys.ruby._cpu_ports[i].slave
if buildEnv['TARGET_ISA'] in ("x86", "arm"):
cpu.itb.walker.port = test_sys.ruby._cpu_ports[i].slave
cpu.dtb.walker.port = test_sys.ruby._cpu_ports[i].slave
if buildEnv['TARGET_ISA'] in "x86":
cpu.interrupts[0].pio = test_sys.ruby._cpu_ports[i].master
cpu.interrupts[0].int_master = test_sys.ruby._cpu_ports[i].slave
cpu.interrupts[0].int_slave = test_sys.ruby._cpu_ports[i].master
else:
if options.caches or options.l2cache:
# By default the IOCache runs at the system clock
test_sys.iocache = IOCache(addr_ranges = test_sys.mem_ranges)
test_sys.iocache.cpu_side = test_sys.iobus.master
test_sys.iocache.mem_side = test_sys.membus.slave
elif not options.external_memory_system:
test_sys.iobridge = Bridge(delay='50ns', ranges = test_sys.mem_ranges)
test_sys.iobridge.slave = test_sys.iobus.master
test_sys.iobridge.master = test_sys.membus.slave
# Sanity check
if options.simpoint_profile:
if not CpuConfig.is_noncaching_cpu(TestCPUClass):
fatal("SimPoint generation should be done with atomic cpu")
if np > 1:
fatal("SimPoint generation not supported with more than one CPUs")
for i in range(np):
if options.simpoint_profile:
test_sys.cpu[i].addSimPointProbe(options.simpoint_interval)
if options.checker:
test_sys.cpu[i].addCheckerCpu()
if options.bp_type:
bpClass = BPConfig.get(options.bp_type)
test_sys.cpu[i].branchPred = bpClass()
if options.indirect_bp_type:
IndirectBPClass = \
BPConfig.get_indirect(options.indirect_bp_type)
test_sys.cpu[i].branchPred.indirectBranchPred = \
IndirectBPClass()
test_sys.cpu[i].createThreads()
# If elastic tracing is enabled when not restoring from checkpoint and
# when not fast forwarding using the atomic cpu, then check that the
# TestCPUClass is DerivO3CPU or inherits from DerivO3CPU. If the check
# passes then attach the elastic trace probe.
# If restoring from checkpoint or fast forwarding, the code that does this for
# FutureCPUClass is in the Simulation module. If the check passes then the
# elastic trace probe is attached to the switch CPUs.
if options.elastic_trace_en and options.checkpoint_restore == None and \
not options.fast_forward:
CpuConfig.config_etrace(TestCPUClass, test_sys.cpu, options)
CacheConfig.config_cache(options, test_sys)
MemConfig.config_mem(options, test_sys)
return test_sys
for i in range(len(host_cpu.workload)):
host_cpu.workload[i].useArchPT = True
host_cpu.workload[i].kvmInSE = True
else:
fatal("KvmCPU can only be used in SE mode with x86")
# configure the TLB hierarchy
GPUTLBConfig.config_tlb_hierarchy(options, system, shader_idx)
# create Ruby system
system.piobus = IOXBar(width=32,
response_latency=0,
frontend_latency=0,
forward_latency=0)
dma_list = [gpu_hsapp, gpu_cmd_proc]
Ruby.create_system(options, None, system, None, dma_list, None)
system.ruby.clk_domain = SrcClockDomain(clock=options.ruby_clock,
voltage_domain=system.voltage_domain)
gpu_cmd_proc.pio = system.piobus.master
gpu_hsapp.pio = system.piobus.master
for i, dma_device in enumerate(dma_list):
exec('system.dma_cntrl%d.clk_domain = system.ruby.clk_domain' % i)
# attach the CPU ports to Ruby
for i in range(options.num_cpus):
ruby_port = system.ruby._cpu_ports[i]
# Create interrupt controller
system.cpu[i].createInterruptController()
#
# create the tester and system, including ruby
#
tester = RubyTester(check_flush=False, checks_to_complete=100, wakeup_frequency=10, num_cpus=options.num_cpus)
# We set the testers as cpu for ruby to find the correct clock domains
# for the L1 Objects.
system = System(cpu=tester)
# Dummy voltage domain for all our clock domains
system.voltage_domain = VoltageDomain(voltage=options.sys_voltage)
system.clk_domain = SrcClockDomain(clock="1GHz", voltage_domain=system.voltage_domain)
system.mem_ranges = AddrRange("256MB")
Ruby.create_system(options, False, system)
# Create a separate clock domain for Ruby
system.ruby.clk_domain = SrcClockDomain(clock="1GHz", voltage_domain=system.voltage_domain)
tester.num_cpus = len(system.ruby._cpu_ports)
#
# The tester is most effective when randomization is turned on and
# artifical delay is randomly inserted on messages
#
system.ruby.randomization = True
for ruby_port in system.ruby._cpu_ports:
#
# Tie the ruby tester ports to the ruby cpu read and write ports
args.l3_assoc = 2
nb_cores = 4
cpus = [TimingSimpleCPU(cpu_id=i) for i in range(nb_cores)]
# overwrite the num_cpus to equal nb_cores
args.num_cpus = nb_cores
# system simulated
system = System(cpu=cpus, clk_domain=SrcClockDomain(clock='1GHz'))
# Create a seperate clock domain for components that should run at
# CPUs frequency
system.cpu.clk_domain = SrcClockDomain(clock='2GHz')
Ruby.create_system(args, False, system)
# Create a separate clock domain for Ruby
system.ruby.clk_domain = SrcClockDomain(clock=args.ruby_clock)
assert (args.num_cpus == len(system.ruby._cpu_ports))
for (i, cpu) in enumerate(system.cpu):
# create the interrupt controller
cpu.createInterruptController()
#
# Tie the cpu ports to the ruby cpu ports
#
cpu.connectAllPorts(system.ruby._cpu_ports[i])
system.voltage_domain = VoltageDomain()
system.clk_domain = SrcClockDomain(clock='1GHz',
voltage_domain=system.voltage_domain)
# Create a seperate clock domain for components that should run at
# CPUs frequency
system.cpu_clk_domain = SrcClockDomain(clock='2GHz',
voltage_domain=system.voltage_domain)
# All cpus are associated with cpu_clk_domain
for cpu in cpus:
cpu.clk_domain = system.cpu_clk_domain
system.mem_ranges = AddrRange('256MB')
Ruby.create_system(options, False, system)
# Create a separate clock domain for Ruby
system.ruby.clk_domain = SrcClockDomain(clock=options.ruby_clock,
voltage_domain=system.voltage_domain)
assert (len(cpus) == len(system.ruby._cpu_ports))
for (i, ruby_port) in enumerate(system.ruby._cpu_ports):
#
# Tie the cpu port to the ruby cpu ports and
# physmem, respectively
#
cpus[i].port = ruby_port.slave
#
if have_kvm_support and buildEnv['TARGET_ISA'] == "x86":
system.vm = KvmVM()
for i in range(len(host_cpu.workload)):
host_cpu.workload[i].useArchPT = True
host_cpu.workload[i].kvmInSE = True
else:
fatal("KvmCPU can only be used in SE mode with x86")
# configure the TLB hierarchy
GPUTLBConfig.config_tlb_hierarchy(args, system, shader_idx)
# create Ruby system
system.piobus = IOXBar(width=32, response_latency=0,
frontend_latency=0, forward_latency=0)
dma_list = [gpu_hsapp, gpu_cmd_proc]
Ruby.create_system(args, None, system, None, dma_list, None)
system.ruby.clk_domain = SrcClockDomain(clock = args.ruby_clock,
voltage_domain = system.voltage_domain)
gpu_cmd_proc.pio = system.piobus.mem_side_ports
gpu_hsapp.pio = system.piobus.mem_side_ports
for i, dma_device in enumerate(dma_list):
exec('system.dma_cntrl%d.clk_domain = system.ruby.clk_domain' % i)
# attach the CPU ports to Ruby
for i in range(args.num_cpus):
ruby_port = system.ruby._cpu_ports[i]
# Create interrupt controller
system.cpu[i].createInterruptController()
system.fpga[0].clk_domain = SrcClockDomain(clock=options.fpga_clock)
system.fpga[0].clk_domain.voltage_domain = VoltageDomain()
system.fpga[0].fpga_bus_addr = 1073741824 * 2
system.fpga[0].size_control_fpga = 29 * 8
system.fpga[0].ModuleName = 'FpgaModule'
system.cpu[0].workload = process1
system.cpu[0].createThreads()
system.piobus = IOXBar()
if options.ruby:
if options.cpu_type == "atomic" or options.cpu_type == "AtomicSimpleCPU":
print >> sys.stderr, "Ruby does not work with atomic cpu!!"
sys.exit(1)
Ruby.create_system(options, False, system, system.piobus)
system.ruby.clk_domain = SrcClockDomain(
clock=options.ruby_clock, voltage_domain=system.voltage_domain)
for i in xrange(np):
print len(system.ruby._cpu_ports)
ruby_port = system.ruby._cpu_ports[i]
system.cpu[i].createInterruptController()
system.cpu[i].icache_port = ruby_port.slave
system.cpu[i].dcache_port = ruby_port.slave
if buildEnv['TARGET_ISA'] == "x86":
system.cpu[i].interrupts[0].pio = ruby_port.master
system.cpu[i].interrupts[0].int_master = ruby_port.slave
system.cpu[i].interrupts[0].int_slave = ruby_port.master
# Create a seperate clock domain for components that should run at
# CPUs frequency
system.cpu[0].clk_domain = SrcClockDomain(clock = '2GHz',
voltage_domain = \
system.voltage_domain)
# configure the TLB hierarchy
GPUTLBConfig.config_tlb_hierarchy(args, system, shader_idx)
# create Ruby system
system.piobus = IOXBar(width=32,
response_latency=0,
frontend_latency=0,
forward_latency=0)
Ruby.create_system(args, None, system)
# Create a separate clock for Ruby
system.ruby.clk_domain = SrcClockDomain(clock=args.ruby_clock,
voltage_domain=system.voltage_domain)
# create the interrupt controller
cpu.createInterruptController()
#
# Tie the cpu cache ports to the ruby cpu ports and
# physmem, respectively
#
cpu.connectAllPorts(system.ruby._cpu_ports[0].in_ports,
system.ruby._cpu_ports[0].in_ports,
system.ruby._cpu_ports[0].interrupt_out_port)