config, arm: Add multi-core KVM support to bL config

Add support for KVM in the big.LITTLE(tm) example configuration. This
replaces the --atomic option with a --cpu-type option that can be used
to switch between atomic, kvm, and timing simulation.

When running in KVM mode, the simulation script automatically assigns
separate event queues (threads) to each of the simulated CPUs. All
simulated devices, including CPU child devices (e.g., interrupt
controllers and caches), are assigned to event queue 0.

Change-Id: Ic9a3f564db91f5a3d3cb754c5a02fdd5c17d5fdf
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-by: Sascha Bischoff <sascha.bischoff@arm.com>
Reviewed-by: Gabor Dozsa <gabor.dozsa@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2561
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Weiping Liao <weipingliao@google.com>

diff --git a/configs/example/arm/devices.py b/configs/example/arm/devices.py
index 7d3f383f309d3955d79aaf1981b09fe3c920c366..f7375cd9a25ac50899ed01c5de9c119ce0ea561b 100644 (file)
--- a/configs/example/arm/devices.py
+++ b/configs/example/arm/devices.py
@@ -1,4 +1,4 @@
-# Copyright (c) 2016 ARM Limited
+# Copyright (c) 2016-2017 ARM Limited
 # All rights reserved.
 #
 # The license below extends only to copyright in the software and shall
@@ -44,6 +44,8 @@ m5.util.addToPath('../../')
 from common.Caches import *
 from common import CpuConfig
 
+have_kvm = "kvm" in CpuConfig.cpu_names()
+
 class L1I(L1_ICache):
     tag_latency = 1
     data_latency = 1
@@ -170,6 +172,14 @@ class AtomicCluster(CpuCluster):
     def addL1(self):
         pass
 
+class KvmCluster(CpuCluster):
+    def __init__(self, system, num_cpus, cpu_clock, cpu_voltage="1.0V"):
+        cpu_config = [ CpuConfig.get("kvm"), None, None, None, None ]
+        super(KvmCluster, self).__init__(system, num_cpus, cpu_clock,
+                                         cpu_voltage, *cpu_config)
+    def addL1(self):
+        pass
+
 
 class SimpleSystem(LinuxArmSystem):
     cache_line_size = 64

diff --git a/configs/example/arm/fs_bigLITTLE.py b/configs/example/arm/fs_bigLITTLE.py
index c2ecf883173d63c93d0803916f620b3dc05107bc..d6825dfaafa37d0e7ed64d7a2d6f688589dedc5b 100644 (file)
--- a/configs/example/arm/fs_bigLITTLE.py
+++ b/configs/example/arm/fs_bigLITTLE.py
@@ -44,6 +44,7 @@ import argparse
 import os
 import sys
 import m5
+import m5.util
 from m5.objects import *
 
 m5.util.addToPath("../../")
@@ -52,6 +53,7 @@ from common import SysPaths
 from common import CpuConfig
 
 import devices
+from devices import AtomicCluster, KvmCluster
 
 
 default_dtb = 'armv8_gem5_v1_big_little_2_2.dtb'
@@ -61,6 +63,21 @@ default_rcs = 'bootscript.rcS'
 
 default_mem_size= "2GB"
 
+def _to_ticks(value):
+    """Helper function to convert a latency from string format to Ticks"""
+
+    return m5.ticks.fromSeconds(m5.util.convert.anyToLatency(value))
+
+def _using_pdes(root):
+    """Determine if the simulator is using multiple parallel event queues"""
+
+    for obj in root.descendants():
+        if not m5.proxy.isproxy(obj.eventq_index) and \
+               obj.eventq_index != root.eventq_index:
+            return True
+
+    return False
+
 
 class BigCluster(devices.CpuCluster):
     def __init__(self, system, num_cpus, cpu_clock,
@@ -107,6 +124,15 @@ def createSystem(caches, kernel, bootscript, disks=[]):
 
     return sys
 
+cpu_types = {
+    "atomic" : (AtomicCluster, AtomicCluster),
+    "timing" : (BigCluster, LittleCluster),
+}
+
+# Only add the KVM CPU if it has been compiled into gem5
+if devices.have_kvm:
+    cpu_types["kvm"] = (KvmCluster, KvmCluster)
+
 
 def addOptions(parser):
     parser.add_argument("--restore-from", type=str, default=None,
@@ -119,8 +145,9 @@ def addOptions(parser):
                         help="Disks to instantiate")
     parser.add_argument("--bootscript", type=str, default=default_rcs,
                         help="Linux bootscript")
-    parser.add_argument("--atomic", action="store_true", default=False,
-                        help="Use atomic CPUs")
+    parser.add_argument("--cpu-type", type=str, choices=cpu_types.keys(),
+                        default="timing",
+                        help="CPU simulation mode. Default: %(default)s")
     parser.add_argument("--kernel-init", type=str, default="/sbin/init",
                         help="Override init")
     parser.add_argument("--big-cpus", type=int, default=1,
@@ -135,9 +162,11 @@ def addOptions(parser):
                         help="Big CPU clock frequency")
     parser.add_argument("--little-cpu-clock", type=str, default="1GHz",
                         help="Little CPU clock frequency")
+    parser.add_argument("--sim-quantum", type=str, default="1ms",
+                        help="Simulation quantum for parallel simulation. " \
+                        "Default: %(default)s")
     return parser
 
-
 def build(options):
     m5.ticks.fixGlobalFrequency()
 
@@ -165,35 +194,31 @@ def build(options):
     root.system = system
     system.boot_osflags = " ".join(kernel_cmd)
 
-    AtomicCluster = devices.AtomicCluster
-
     if options.big_cpus + options.little_cpus == 0:
         m5.util.panic("Empty CPU clusters")
 
+    big_model, little_model = cpu_types[options.cpu_type]
+
+    all_cpus = []
     # big cluster
     if options.big_cpus > 0:
-        if options.atomic:
-            system.bigCluster = AtomicCluster(system, options.big_cpus,
-                                              options.big_cpu_clock)
-        else:
-            system.bigCluster = BigCluster(system, options.big_cpus,
-                                           options.big_cpu_clock)
-        mem_mode = system.bigCluster.memoryMode()
+        system.bigCluster = big_model(system, options.big_cpus,
+                                      options.big_cpu_clock)
+        system.mem_mode = system.bigCluster.memoryMode()
+        all_cpus += system.bigCluster.cpus
+
     # little cluster
     if options.little_cpus > 0:
-        if options.atomic:
-            system.littleCluster = AtomicCluster(system, options.little_cpus,
-                                                 options.little_cpu_clock)
+        system.littleCluster = little_model(system, options.little_cpus,
+                                            options.little_cpu_clock)
+        system.mem_mode = system.littleCluster.memoryMode()
+        all_cpus += system.littleCluster.cpus
 
-        else:
-            system.littleCluster = LittleCluster(system, options.little_cpus,
-                                                 options.little_cpu_clock)
-        mem_mode = system.littleCluster.memoryMode()
+    # Figure out the memory mode
+    if options.big_cpus > 0 and options.little_cpus > 0 and \
+       system.littleCluster.memoryMode() != system.littleCluster.memoryMode():
+        m5.util.panic("Memory mode missmatch among CPU clusters")
 
-    if options.big_cpus > 0 and options.little_cpus > 0:
-        if system.bigCluster.memoryMode() != system.littleCluster.memoryMode():
-            m5.util.panic("Memory mode missmatch among CPU clusters")
-    system.mem_mode = mem_mode
 
     # create caches
     system.addCaches(options.caches, options.last_cache_level)
@@ -203,13 +228,43 @@ def build(options):
         if options.little_cpus > 0 and system.littleCluster.requireCaches():
             m5.util.panic("Little CPU model requires caches")
 
+    # Create a KVM VM and do KVM-specific configuration
+    if issubclass(big_model, KvmCluster):
+        _build_kvm(system, all_cpus)
+
     # Linux device tree
     system.dtb_filename = SysPaths.binary(options.dtb)
 
     return root
 
+def _build_kvm(system, cpus):
+    system.kvm_vm = KvmVM()
+
+    # 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.
+    if len(cpus) > 1:
+        device_eq = 0
+        first_cpu_eq = 1
+        for idx, cpu in enumerate(cpus):
+            # 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 = device_eq
+            cpu.eventq_index = first_cpu_eq + idx
+
+
 
 def instantiate(options, checkpoint_dir=None):
+    # Setup the simulation quantum if we are running in PDES-mode
+    # (e.g., when using KVM)
+    root = Root.getInstance()
+    if root and _using_pdes(root):
+        m5.util.inform("Running in PDES mode with a %s simulation quantum.",
+                       options.sim_quantum)
+        root.sim_quantum = _to_ticks(options.sim_quantum)
+
     # Get and load from the chkpt or simpoint checkpoint
     if options.restore_from:
         if checkpoint_dir and not os.path.isabs(options.restore_from):