gem5.git
7 years agostats: Add a boat load of stats to the SPARC solaris boot regression.
Gabe Black [Tue, 4 Apr 2017 07:56:29 +0000 (00:56 -0700)]
stats: Add a boat load of stats to the SPARC solaris boot regression.

A large number of stats were added by the following change:

    commit 5350879f499470a2683dfec6cff021dd7ac20fa6
    Author: David Guillen Fandos <david.guillen@arm.com>
    Date:   Mon Jun 6 17:16:43 2016 +0100

        pwr: Add power states to ClockedObject

Change-Id: Iec32bb7f701db0a09be26fe5ffb2812385f972c2
Reviewed-on: https://gem5-review.googlesource.com/2642
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
7 years agostats: Un-empty the SPARC FS stats.txt file.
Gabe Black [Tue, 28 Mar 2017 06:51:10 +0000 (23:51 -0700)]
stats: Un-empty the SPARC FS stats.txt file.

This was emptied accidentally by the CL below. A lot of other files were too,
but those were eventually refilled.

    commit 62b6ff22ec1f90014b1d0fc778014bdb38cc09ce
    Author: Curtis Dunham <Curtis.Dunham@arm.com>
    Date:   Tue May 31 11:07:18 2016 +0100

        stats: update for snoop filter tweak

Change-Id: I34aefca51a92a6a98f6a8fdbdab7106cc1fff171
Reviewed-on: https://gem5-review.googlesource.com/2641
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>

7 years agoscons: Fix hook installation error caused by stale cache
Andreas Sandberg [Wed, 5 Apr 2017 08:53:43 +0000 (09:53 +0100)]
scons: Fix hook installation error caused by stale cache

Due to the way SCons caches some file system state internally, it
sometimes "remembers" that a file or directory didn't exist at some
point. The git hook installation script sometimes needs to create a
hooks directory in the repository. Due to the cached state of the
hooks directory, the build system tries to create it twice. The second
mkdir operation leads to an error since the directory already exists.

Fix this issue by clearing the cached state of the hooks directory
after creating it.

Change-Id: I3f67f75c06ef928b439a0742f7619b7e92ed093b
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2660
Reviewed-by: Gabe Black <gabeblack@google.com>
7 years agoarm, kvm: implement GIC state transfer
Curtis Dunham [Fri, 27 Jan 2017 20:21:59 +0000 (20:21 +0000)]
arm, kvm: implement GIC state transfer

This also allows checkpointing of a Kvm GIC via the Pl390 model.

Change-Id: Ic85d81cfefad630617491b732398f5e6a5f34c0b
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2444
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Weiping Liao <weipingliao@google.com>
7 years agoarm, dev: add basic support for GICC_BPR register
Curtis Dunham [Thu, 9 Mar 2017 17:30:59 +0000 (17:30 +0000)]
arm, dev: add basic support for GICC_BPR register

The Binary Point Register (BPR) specifies which bits belong to the
group priority field (which are used for preemption) and which to the
subpriority field (which are ignored for preemption).

Change-Id: If51e669d23b49047b69b82ab363dd01a936cc93b
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2443
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Weiping Liao <weipingliao@google.com>
7 years agoarm, dev: refactor GIC Pl390 GICD_ITARGETSRn handling
Curtis Dunham [Tue, 31 Jan 2017 17:11:24 +0000 (17:11 +0000)]
arm, dev: refactor GIC Pl390 GICD_ITARGETSRn handling

The aforementioned registers (Interrupt Processor Targets Registers) are
banked per-CPU, but are read-only.  This patch eliminates the per-CPU
storage of these values that are simply computed.

Change-Id: I52cafc2f58e87dd54239a71326c01f4923544689
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2442
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Weiping Liao <weipingliao@google.com>
7 years agoarm: refactor packet processing in Pl390 GIC
Curtis Dunham [Fri, 27 Jan 2017 20:16:23 +0000 (20:16 +0000)]
arm: refactor packet processing in Pl390 GIC

Change-Id: I696703418506522ba90df5c2c4ca45c95a6efbea
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2441
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Weiping Liao <weipingliao@google.com>
7 years agoarm: Don't panic when checking coprocessor read/write permissions
Nikos Nikoleris [Wed, 1 Mar 2017 20:55:15 +0000 (20:55 +0000)]
arm: Don't panic when checking coprocessor read/write permissions

Instructions that use the coprocessor interface check the current
program status to determine whether the current context has the
priviledges to read from/write to the coprocessor. Some modes allow
the execution of coprocessor instructions, some others do not allow it,
while some other modes are unexpected (e.g., executing an AArch32
instruction while being in an AArch64 mode).

Previously we would unconditionally trigger a panic if we were in an
unexpected mode. This change removes the panic and replaces it
with an Undefined Instruction fault that triggers if and when a
coprocessor instruction commits in an unexpected mode. This allows
speculative coprocessor instructions from unexpected modes to execute
but prevents them from gettting committed.

Change-Id: If2776d5bae2471cdbaf76d0e1ae655f501bfbf01
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Rekai Gonzalez Alberquilla <rekai.gonzalezalberquilla@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2281
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Weiping Liao <weipingliao@google.com>
7 years agoarm: Treat Write-Through Normal memory as Non-Cacheable
Nikos Nikoleris [Mon, 27 Feb 2017 15:57:31 +0000 (15:57 +0000)]
arm: Treat Write-Through Normal memory as Non-Cacheable

A completed write to a memory location that is Write-Through Cacheable
has to be visible to an external observer without the need of explicit
cache maintenance. This change adds support for Write-Through
Cacheable Normal memory and treats it as Non-cacheable. This incurs a
small penalty as accesses to the memory do not fill in the cache but
does not violate the properties of the memory type.

Change-Id: Iee17ef9d952a550be9ad660b1e60e9f6c4ef2c2d
Reviewed-by: Giacomo Gabrielli <giacomo.gabrielli@arm.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2280
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
7 years agosim: Handle cases where Drainable::resume() creates objects
Andreas Sandberg [Tue, 28 Mar 2017 13:50:06 +0000 (13:50 +0000)]
sim: Handle cases where Drainable::resume() creates objects

There are cases where Drainable objects need to create new objects in
Drainable::resume(). In such cases, the local drain state will be
inherited from the DrainManager. We currently set the state to Running
as soon as we start resuming the simulator. This means that new
objects are created in the Running state rather than the Drained
state, which the resume code assumes. Depending on the traversal order
in DrainManager::resume(), this sometimes triggers a panic because the
object being resumed is in the wrong state.

This change introduces a new drain state, Resuming, that the
DrainManager enters as soon as it starts resuming the
simulator. Objects that are created while resuming are created in this
state. Such objects are then resumed in a subsequent pass over the
list of Drainable objects that need to be resumed. Once all objects
have been resumed, the simulator enters the Running state.

Change-Id: Ieee8645351ffbdec477e9cd2ff86fc795e459617
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2600
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Weiping Liao <weipingliao@google.com>
7 years agoconfig, arm: Add multi-core KVM support to bL config
Andreas Sandberg [Fri, 17 Mar 2017 11:10:55 +0000 (11:10 +0000)]
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>
7 years agoconfig, arm: Unify checkpoint path handling in bL configs
Andreas Sandberg [Mon, 20 Mar 2017 10:44:07 +0000 (10:44 +0000)]
config, arm: Unify checkpoint path handling in bL configs

The vanilla bL configuration file and the dist-gem5 configuration file
use slightly different code paths when restoring from
checkpoints. Unify this by passing the parsed options to the
instantiate() method and adding an optional checkpoint keyword
argument for checkpoint directories (only used by the dist-gem5
script).

Change-Id: I9943ec10bd7a256465e29c8de571142ec3fbaa0e
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/2560
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Weiping Liao <weipingliao@google.com>
7 years agoarm, kvm: Override the kernel's default MPIDR value
Andreas Sandberg [Mon, 20 Mar 2017 14:36:48 +0000 (14:36 +0000)]
arm, kvm: Override the kernel's default MPIDR value

The kernel and gem5 derive MPIDR values from CPU IDs in slightly
different ways. This means that guests running in a multi-CPU setup
sometimes fail to bring up secondary CPUs. Fix this by overriding the
MPIDR value in virtual CPUs just after they have been instantiated.

Change-Id: I916d44978a9c855ab89c80a083af45b0cea6edac
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-on: https://gem5-review.googlesource.com/2461
Reviewed-by: Weiping Liao <weipingliao@google.com>
7 years agodev, arm: Fix multi-core KVM race in the generic timer
Andreas Sandberg [Fri, 17 Mar 2017 11:57:23 +0000 (11:57 +0000)]
dev, arm: Fix multi-core KVM race in the generic timer

The generic timer sometimes needs to access global state. This can
lead to race conditions when simulating a multi-core KVM system where
each core lives in its own thread. In that case, the setMiscReg and
readMiscReg methods are called from the thread owning the CPU and not
the global device thread.

Change-Id: Ie3e982258648c8562cce0b30a0c122dfbfaf42cd
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Gabor Dozsa <gabor.dozsa@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-by: Sascha Bischoff <sascha.bischoff@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2460
Reviewed-by: Weiping Liao <weipingliao@google.com>
7 years agodev: Align BAR0 size to power of 2 for VirtIO devices
Sascha Bischoff [Fri, 24 Mar 2017 16:02:30 +0000 (16:02 +0000)]
dev: Align BAR0 size to power of 2 for VirtIO devices

When setting the size of a PCI BAR, the kernel only supports powers of
two (as per the PCI spec). Previously, the size was incorrectly read
by the kernel, and the address ranges assigned to the PCI devices
could overlap, resulting in gem5 crashes.  We now round up to the next
power of two.

Kudos to Sergei Trofimov who helped to debug this issue!

Change-Id: I54ca399b62ea07c09d4cd989b17dfa670e841bbe
Reviewed-by: Anouk Van Laer <anouk.vanlaer@arm.com>
Reviewed-by: Sergei Trofimov <sergei.trofimov@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2580
Reviewed-by: Paul Rosenfeld <prosenfeld@micron.com>
7 years agodev: Add a dummy VirtIO device
Andreas Sandberg [Mon, 7 Nov 2016 18:16:51 +0000 (18:16 +0000)]
dev: Add a dummy VirtIO device

VirtIO transport interfaces always expect a VirtIO device
pointer. However, there are cases (in particular when using VirtIO's
MMIO interface) where we want to instantiate an interface without a
device. Add a dummy device using VirtIO device ID 0 and no queues to
handle this use case.

Change-Id: I6cbe12fd403903ef585be40279c3b1321fde48ff
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Sudhanshu Jha <sudhanshu.jha@arm.com>
Reviewed-by: Rekai Gonzalez Alberquilla <rekai.gonzalezalberquilla@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2325
Reviewed-by: Weiping Liao <weipingliao@google.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
7 years agodev: Rename VirtIO PCI debug flag
Andreas Sandberg [Mon, 7 Nov 2016 18:14:50 +0000 (18:14 +0000)]
dev: Rename VirtIO PCI debug flag

Rename VIOPci -> VIOIface to avoid having a separate flag for the MMIO
interface.

Change-Id: I99f9210fa36ce33662c48537fd3992cd9a69d349
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Sudhanshu Jha <sudhanshu.jha@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Rekai Gonzalez Alberquilla <rekai.gonzalezalberquilla@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2324
Reviewed-by: Weiping Liao <weipingliao@google.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
7 years agoarm: fix template instantiation warning in clang
Matteo Andreozzi [Mon, 3 Apr 2017 12:38:58 +0000 (13:38 +0100)]
arm: fix template instantiation warning in clang

In arch/arm/faults.hh, template the static member vals require explicit
specialisation to avoid compiler warnings.

Change-Id: Ie404ccaa43269cb1bb819e33153e776abbf3a79b
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
7 years agoscons: Collapse symlinks when installing git hooks.
Gabe Black [Sat, 25 Mar 2017 09:40:44 +0000 (02:40 -0700)]
scons: Collapse symlinks when installing git hooks.

When naively computing the relative path between the git hooks directory and a
hook we want to install, that will generally start with a few ".." path
components to work back out of the .git directory into the working directory.'
If the hooks directory is actually a symlink, then following ".." directory
entries won't get us back to where we came from, they'll take us to the actual
parent directory of hooks. The relative path we computed would then try to go
forward from this other directory using a path that would have worked in the
working directory, hopefully going somewhere that doesn't exist, but
potentially going to a totally unrelated file with the same relative path.

To avoid this problem, we should expand any symlinks in both the hooks
directory path, and the path to the hook script. That way, any ".." components
will go where we'd expect them to, and the relative path will actually go from
hooks to the script we expect.

Change-Id: I64d51bc817351f89b1d60eceaf450cc0a4553415
Reviewed-on: https://gem5-review.googlesource.com/2542
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>

7 years agoscons: Use a relative symlink for git hooks more selectively.
Gabe Black [Sat, 25 Mar 2017 09:26:04 +0000 (02:26 -0700)]
scons: Use a relative symlink for git hooks more selectively.

If the hooks directory is a symlink, then there are at least two possible
scenarios to consider when installing a hook which is itself a symlink. The
first is that hooks is a relative symlink, and so is likely intended to stay
in place relative to .git and the git working directory. In that case, it's ok
for the symlinks inside of hooks to be relative to the working directory too,
since they should also stay in place relatively speaking.

The second situation is that the symlink is absolute. In that case, moving the
git working directory will move the hook relative to the hook directory, and
any relative symlink will become broken. In that case, the hook symlink needs
to be absolute.

The same logic likely applies to the .git directory itself, although I haven't
run into a situation in practice where the .git directory is actually a
symlink.

Change-Id: I047aa198094dd0fd5a841417d93b211ece02783f
Reviewed-on: https://gem5-review.googlesource.com/2541
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>

7 years agoscons: Detect and remove broken git hook symlinks.
Gabe Black [Sat, 25 Mar 2017 09:06:39 +0000 (02:06 -0700)]
scons: Detect and remove broken git hook symlinks.

When installing a git hook, it's possible for the hook to not "exist" if it's
actually a symlink which points to a file that doesn't exist. Trying to create
a new symlink in its place without first removing the old one causes a build
failure in these cases.

If the hook doesn't "exist" but is still a link, that means it's actually a
broken link and should be deleted by the hook installation function before any
new symlink is created.

Change-Id: I59aa51feb5bd74ca33e51e89cde2ceabeb41bd76
Reviewed-on: https://gem5-review.googlesource.com/2540
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>

7 years agoscons: Stop generating an a.out checking the "as" version.
Gabe Black [Fri, 24 Mar 2017 07:36:16 +0000 (00:36 -0700)]
scons: Stop generating an a.out checking the "as" version.

Change-Id: I71d07fc64bdb3c6c3e93e2a1fd358cc899a70678
Reviewed-on: https://gem5-review.googlesource.com/2500
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>

7 years agoarm: correct register read bug in Pl390 GIC
Curtis Dunham [Fri, 20 Jan 2017 20:55:03 +0000 (20:55 +0000)]
arm: correct register read bug in Pl390 GIC

Change-Id: I4c0de7c2a5b40c1a9f009ca12062cb108b450b04
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
7 years agopython: Automatically disable listeners in batch setups
Andreas Sandberg [Mon, 28 Nov 2016 16:40:45 +0000 (16:40 +0000)]
python: Automatically disable listeners in batch setups

Determine if gem5 is running in a batch environment by checking if
STDIN is wired to a TTY or not. If the simulator is running in a batch
environment, disable all listeners by default. This behavior can be
overridden using the --enable-listeners option.

Change-Id: I404c709135339144216bf08a2769c016c543333c
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Sean McGoogan <sean.mcgoogan@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2322
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
7 years agoutil: Add a tool to list outgoing/incoming changes
Andreas Sandberg [Tue, 7 Mar 2017 14:38:25 +0000 (14:38 +0000)]
util: Add a tool to list outgoing/incoming changes

Add a small Python script that uses Gerrit's Change-Id: tags to list
incoming and outgoing changes.

Change-Id: Iea1757b2d64a57a4c7b4e47718cfcaa725a99615
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Sascha Bischoff <sascha.bischoff@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2329
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>

7 years agomisc: add copyright/name information for contribution
Pierre-Yves Péneau [Fri, 10 Mar 2017 08:58:16 +0000 (09:58 +0100)]
misc: add copyright/name information for contribution

Change-Id: I9242ce50b86b02ec1880d411627da11265cb8961
Signed-off-by: Pierre-Yves Péneau <pierre-yves.peneau@lirmm.fr>
Reviewed-on: https://gem5-review.googlesource.com/2328
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Brandon Potter <Brandon.Potter@amd.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>

7 years agosyscall-emul: Hotfix for FreeBSD/Mac builds
Brandon Potter [Sun, 29 Jan 2017 12:22:33 +0000 (12:22 +0000)]
syscall-emul: Hotfix for FreeBSD/Mac builds

The clone system call added in 236719892 relies on header files
from Linux systems. Obviously, this prevents compilation for
anyone using FreeBSD or Mac to compile the simulator. This
changeset is meant as a temporary fix to allow builds on
non-Linux systems until a proper solution is found.

Change-Id: I404cc41c588ed193dd2c1ca0c1aea35b0786fe4e
Reviewed-on: https://gem5-review.googlesource.com/2420
Maintainer: Brandon Potter <Brandon.Potter@amd.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
7 years agosyscall-emul: change NULL to nullptr in Process files
Brandon Potter [Wed, 15 Mar 2017 16:09:36 +0000 (11:09 -0500)]
syscall-emul: change NULL to nullptr in Process files

Change-Id: I9ff21092876593237f919e9f7fb7283bd865ba2e
Reviewed-on: https://gem5-review.googlesource.com/2421
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Brandon Potter <Brandon.Potter@amd.com>

7 years agocpu: Print progress messages in Trace CPU
Radhika Jagtap [Tue, 16 Aug 2016 13:14:58 +0000 (14:14 +0100)]
cpu: Print progress messages in Trace CPU

This change adds the ability to print a message at intervals
of committed instruction count to indicate progress in the
trace replay.

Change-Id: I8363502354c42bfc52936d2627986598b63a5797
Reviewed-by: Rekai Gonzalez Alberquilla <rekai.gonzalezalberquilla@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2321
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
7 years agotests: Warn not fail when reading invalid pickle status files
Nikos Nikoleris [Thu, 1 Sep 2016 16:27:20 +0000 (17:27 +0100)]
tests: Warn not fail when reading invalid pickle status files

With this change, the test script will output a warning when it reads
an incomplete (e.g., when a regression is still running) or corrupt
status file instead of throwing an exception. When the scipt is used
to show the results the corrupt file is skipped; when it is used to
test if all regressions run successfully it will return an error value
(2).

Change-Id: Ie7d9b457b200e3abc7ae6238e3efbf3d18cf4297
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2320
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>

7 years agoarm, dev: Add missing override in the Pl390 GIC model
Andreas Sandberg [Wed, 15 Mar 2017 14:36:37 +0000 (14:36 +0000)]
arm, dev: Add missing override in the Pl390 GIC model

The Pl390::getAddrRanges() method should have been flagged using the
override keyword. Other methods in this class already use the override
keyword, so this results in a warning about inconsistent override
usage when compiling using clang.

Change-Id: I17449687a8e074262232562487b58c96466bd54e
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
7 years agodev, arm: Add draining to the GIC model
Andreas Sandberg [Wed, 17 Feb 2016 14:50:41 +0000 (08:50 -0600)]
dev, arm: Add draining to the GIC model

The GIC model currently adds a delay to interrupts when posting them
to a target CPU. This means that an interrupt signal will be
represented by an event for a short period of time. We currently
ignore this when draining and serialize the tick when the interrupt
will fire. Upon loading the checkpoint, the simulated GIC reschedules
the pending events. This behaviour is undesirable when we implement
support for switching between in-kernel GIC emulation and gem5 GIC
emulation. In that case, the (kernel) GIC model gets a lot simpler if
we don't need to worry about in-flight interrupts from the gem5 GIC.

This changeset adds a draining check to force the GIC into a state
where all interrupts have been delivered prior to checkpointing/CPU
switching. It also removes the now redundant serialization of
interrupt events.

Change-Id: I8b8b080aa291ca029a3a7bdd1777f1fcd5b01179
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2331
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
7 years agoarm: Clean up the GIC implementation
Andreas Sandberg [Tue, 27 Sep 2016 13:16:50 +0000 (14:16 +0100)]
arm: Clean up the GIC implementation

Lots of minor cleaups:
  * Make cached params const
  * Don't serialize params
  * Use AddrRange to represent the distributor and CPU address spaces
  * Store a const AddrRangeList of all PIO ranges

Change-Id: I40a17bc3a38868fb3b8af247790e852cf99ddf1d
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2330
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
7 years agostyle: change NULL to nullptr in syscall files
Brandon Potter [Wed, 1 Mar 2017 21:15:51 +0000 (15:15 -0600)]
style: change NULL to nullptr in syscall files

Change-Id: I02719f3572f6665cace1eb5681f297dcde9e71ce
Reviewed-on: https://gem5-review.googlesource.com/2271
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Michael LeBeane <Michael.Lebeane@amd.com>
Reviewed-by: Tony Gutierrez <anthony.gutierrez@amd.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>

7 years agosyscall-emul: Ignore unimplemented system calls
Brandon Potter [Wed, 1 Mar 2017 21:11:50 +0000 (15:11 -0600)]
syscall-emul: Ignore unimplemented system calls

This changeset sets the implementation policy for a subset of
system calls to the ignoreFunc implementation (for x86 only).
The ignored system calls likely will never be implemented and
this allows a warning to be issued instead of the simulation
exiting with a fatal.

Change-Id: I8d9741ad683151e88cc71156d3602e2d0ccb0acf
Reviewed-on: https://gem5-review.googlesource.com/2270
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Tony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Michael LeBeane <Michael.Lebeane@amd.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
7 years agosyscall-emul: Rewrite system call exit code
Brandon Potter [Wed, 1 Mar 2017 20:52:23 +0000 (14:52 -0600)]
syscall-emul: Rewrite system call exit code

The changeset does a major refactor on the exit, exit_group, and
futex system calls regarding exit functionality.

A FutexMap class and related structures are added into a new
file. This increases code clarity by encapsulating the futex
operations and the futex state into an object.

Several exit conditions were added to allow the simulator to end
processes under certain conditions. Also, the simulation only
exits now when all processes have finished executing.

Change-Id: I1ee244caa9b5586fe7375e5b9b50fd3959b9655e
Reviewed-on: https://gem5-review.googlesource.com/2269
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
7 years agosyscall-emul: Add the tgkill system call
Brandon Potter [Wed, 1 Mar 2017 20:34:11 +0000 (14:34 -0600)]
syscall-emul: Add the tgkill system call

This changeset adds support to kill a thread group by calling
the tgkill system call. The functionality is needed in some
pthread applications.

Change-Id: I0413a3331be69b74dfab30de95384113ec4efb63
Reviewed-on: https://gem5-review.googlesource.com/2268
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Tony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Michael LeBeane <Michael.Lebeane@amd.com>
7 years agosyscall-emul: Adds SE mode signal feature
Brandon Potter [Wed, 1 Mar 2017 20:29:00 +0000 (14:29 -0600)]
syscall-emul: Adds SE mode signal feature

This changeset adds a simple class definition and a member
in the System object to track signals sent between processes.
The implementation cannot support all signals that might be
sent between processes, but it can support some of the simple
use cases like SIGCHLD.

Change-Id: Id5f95aa60e7f49da1c5b5596fbfa26e729453ac7
Reviewed-on: https://gem5-review.googlesource.com/2267
Reviewed-by: Michael LeBeane <Michael.Lebeane@amd.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>

7 years agosyscall-emul: Add or extend dup, dup2, and pipe
Brandon Potter [Wed, 1 Mar 2017 19:35:02 +0000 (13:35 -0600)]
syscall-emul: Add or extend dup, dup2, and pipe

This changeset extends the pipe system call to work with
architectures other than Alpha (and enables the syscall for
x86). For the dup system call, it sets the clone-on-exec
flag by default. For the dup2 system call, the changeset
adds an implementation (and enables it for x86).

Change-Id: I00ddb416744ee7dd61a5cd02c4c3d97f30543878
Reviewed-on: https://gem5-review.googlesource.com/2266
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Tony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Michael LeBeane <Michael.Lebeane@amd.com>
7 years agosyscall-emul: Add functionality to open syscalls
Brandon Potter [Wed, 1 Mar 2017 19:24:16 +0000 (13:24 -0600)]
syscall-emul: Add functionality to open syscalls

This changeset adds refactors the existing open system call,
adds the openat variant (enabled for x86 builds), and adds
additional "special file" test cases for /proc/meminfo and
/etc/passwd.

Change-Id: I6f429db65bbf2a28ffa3fd12df518c2d0de49663
Reviewed-on: https://gem5-review.googlesource.com/2265
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Tony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Michael LeBeane <Michael.Lebeane@amd.com>
7 years agostyle: Correct some style issues
Brandon Potter [Wed, 1 Mar 2017 19:18:49 +0000 (13:18 -0600)]
style: Correct some style issues

This changeset fixes line alignment issues, spacing, spelling,
etc. for files that are used during SE Mode.

Change-Id: Ie61b8d0eb4ebb5af554d72f1297808027833616e
Reviewed-on: https://gem5-review.googlesource.com/2264
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Tony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Michael LeBeane <Michael.Lebeane@amd.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Pierre-Yves Péneau <pierre-yves.peneau@lirmm.fr>
7 years agosyscall-emul: Move memState into its own file
Brandon Potter [Wed, 1 Mar 2017 19:07:43 +0000 (13:07 -0600)]
syscall-emul: Move memState into its own file

The Process class is full of implementation details and
structures related to SE Mode. This changeset factors out an
internal class from Process and moves it into a separate file.
The purpose behind doing this is to clean up the code and make
it a bit more modular.

Change-Id: Ic6941a1657751e8d51d5b6b1dcc04f1195884280
Reviewed-on: https://gem5-review.googlesource.com/2263
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>

7 years agomisc: add missing copyright/author information in previous commit
Pierre-Yves Péneau [Fri, 24 Feb 2017 08:45:21 +0000 (09:45 +0100)]
misc: add missing copyright/author information in previous commit

See a06a46f and a854373.

Change-Id: Id66427db22b7d7764c218b9cd78d95db929f4127
Signed-off-by: Pierre-Yves Péneau <pierre-yves.peneau@lirmm.fr>
Reviewed-on: https://gem5-review.googlesource.com/2224
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>

7 years agoruby: fix MOESI_hammer directory to work with > 3GB memory
Lena Olson [Sun, 5 Feb 2017 23:20:34 +0000 (17:20 -0600)]
ruby: fix MOESI_hammer directory to work with > 3GB memory

The MOESI_hammer directory assumes a contiguous address space, but X86
has an IO gap from 3-4GB. This patch allows the directory to work with
more than 3GB of memory on X86.

Assumptions: the physical address space (range of possible physical
addresses) is 0-XGB when X <= 3GB, and 0-(X+1)GB when X > 3GB. If there
is no IO gap this patch should still work.

Change-Id: I5453a09e953643cada2c096a91d339a3676f55ee
Reviewed-on: https://gem5-review.googlesource.com/2169
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>

7 years agoruby: fix and/or precedence in slicc
Lena Olson [Sun, 5 Feb 2017 20:47:37 +0000 (14:47 -0600)]
ruby: fix and/or precedence in slicc

The slicc compiler currently treats && and || with the same precedence.
This is highly non-intuitive to people used to C, and was probably an
error. This patch makes && bind tighter than ||.

For example, previously:
if (A || B && C)
compiled to:
if ((A || B) && C)
With this patch, it compiles to:
if (A || (B && C))

Change-Id: Idbbd5b50cc86a8d6601045adc14a253284d7b791
Signed-off-by: Lena Olson (leolson@google.com)
Reviewed-on: https://gem5-review.googlesource.com/2168
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Joe Gross <criusx@gmail.com>
Reviewed-by: Sooraj Puthoor <puthoorsooraj@gmail.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>

Conflicts:
COPYING

7 years agomisc: Add a CONTRIBUTING document
Jason Lowe-Power [Thu, 9 Mar 2017 16:13:10 +0000 (10:13 -0600)]
misc: Add a CONTRIBUTING document

This document details how to contribute to gem5 based on our new
contribution flow with git and gerrit.

Change-Id: I0a7e15fd83a3ee3ab6c85c1192f46f1e1d33b7c2
Signed-off-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-on: http://reviews.gem5.org/r/3814/
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Tony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Pierre-Yves Peneau <pierre-yves.peneau@lirmm.fr>
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
Reviewed-by: Brad Beckmann <brad.beckmann@amd.com>
Reviewed-by: Ali Saidi <Ali.Saidi@ARM.com>
7 years agogpu-compute: Fix Python/C++ object hierarchy discrepancies
Andreas Sandberg [Mon, 27 Feb 2017 13:17:51 +0000 (13:17 +0000)]
gpu-compute: Fix Python/C++ object hierarchy discrepancies

The GPUCoalescer and the Shader classes have different base classes in
C++ and Python. This causes subtle bugs in SWIG and compilation errors
for PyBind.

Change-Id: I1ddd2a8ea43f083470538ddfea891347b21d14d8
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2228
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Tony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Pierre-Yves Péneau <pierre-yves.peneau@lirmm.fr>
Reviewed-by: Bradford Beckmann <brad.beckmann@amd.com>
7 years agopower: Avoid forward declarations that confuse wrappers
Andreas Sandberg [Mon, 27 Feb 2017 13:17:51 +0000 (13:17 +0000)]
power: Avoid forward declarations that confuse wrappers

The Python wrappers get confused by the forward declarations in the
power framework. This changeset restructures the code slightly to
avoid the troublesome forward declarations.

Change-Id: Id8c93224f1988edb5fdf9d3abc6237f2f688c02d
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-by: Sascha Bischoff <sascha.bischoff@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2227
Reviewed-by: Pierre-Yves Péneau <pierre-yves.peneau@lirmm.fr>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>

7 years agodev, arm: Render HDLCD frames at a fixed rate in KVM
Sudhanshu Jha [Wed, 1 Mar 2017 17:38:51 +0000 (17:38 +0000)]
dev, arm: Render HDLCD frames at a fixed rate in KVM

Use the new fast scan-out API in the PixelPump to render frames at a
fixed frame rate in KVM mode. The refresh rate when running in KVM can
be controlled by the virt_refresh_rate parameter.

Change-Id: Ib3c78f174e3f8f4ca8a9b723c4e5d311a433b8aa
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2242
Reviewed-by: Rahul Thakur <rjthakur@google.com>
7 years agodev: Add support for single-pass scan out in the PixelPump
Sudhanshu Jha [Fri, 24 Feb 2017 13:34:22 +0000 (13:34 +0000)]
dev: Add support for single-pass scan out in the PixelPump

Add a helper function to scan out an entire frame in one time
step. This requires the public PixelPump to be changed somewhat to
separate timing updates from general PixelPump control. Instead of
calling PixelPump::start(timings), timings now need to be updated
using a separate call to PixelPump::updateTimings(timings) before
calling PixelPump::start().

Display controllers that don't need accurate timing (e.g., in KVM
mode), can use the new PixelPump::renderFrame() API to render an
entire frame in one step. This call results in the same callbacks
(e.g., calls to nextPixel()) as the timing calls, but they all happen
in immediately. Unlike the timing counterpart, renderFrame() doesn't
support buffer underruns and will panic if nextPixle() indicates an
underrun.

Change-Id: I76c84db04249b02d4207c5281d82aa693d0881be
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2241
Reviewed-by: Rahul Thakur <rjthakur@google.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
7 years agodev, kvm: Add a fast KVM-aware mode in DmaReadFifo
Sudhanshu Jha [Mon, 27 Feb 2017 10:29:56 +0000 (10:29 +0000)]
dev, kvm: Add a fast KVM-aware mode in DmaReadFifo

Use a fast, functional, read operations keep the DMA FIFO full when
running in KVM mode.

Change-Id: I5b378c2fb6a1d3e687cef15e807e63a0a53a60e2
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2226
Reviewed-by: Rahul Thakur <rjthakur@google.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>

7 years agoarm, kmi: Clear interrupts in KMI devices
Sudhanshu Jha [Mon, 27 Feb 2017 10:29:56 +0000 (10:29 +0000)]
arm, kmi: Clear interrupts in KMI devices

Added functionality to check and clear interrupts for KMI
devices. This fixes a boot bug when using KVM and in-kernel GIC
emulation.

Change-Id: Ia3e91d07567b7faf3f82b0adfda4a165a502a339
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2225
Reviewed-by: Rahul Thakur <rjthakur@google.com>
7 years agosyscall-emul: Remove unused class and member
Brandon Potter [Wed, 1 Mar 2017 17:10:28 +0000 (11:10 -0600)]
syscall-emul: Remove unused class and member

The WaitRec structure in the Process class is unnecessary. There
is a member declaration inside of the Process class, waitList,
that uses the WaitRec definition. However, waitList is unused so
they are both dead bits of code. This changeset removes both the
WaitRec struct and waitList member from Process.

Change-Id: Ia6ee7488b9f47fd0f0ae29c818fba6ea0710699c
Reviewed-on: https://gem5-review.googlesource.com/2262
Reviewed-by: Tony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Michael LeBeane <Michael.Lebeane@amd.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>

7 years agomem: Make blkAlign a common function between all tag classes
Nikos Nikoleris [Mon, 31 Oct 2016 12:02:24 +0000 (12:02 +0000)]
mem: Make blkAlign a common function between all tag classes

blkAlign was defined as a separate function in the base associative
and fully-associative tags classes although both functions implemented
identical functionality. This patch moves the blkAlign in the base
tags class.

Change-Id: I3d415d0e62bddeec7ce0d559667e40a8c5fdc2d4
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
7 years agomem: Use pkt::getBlockAddr instead of BaseCace::blockAlign
Nikos Nikoleris [Wed, 26 Oct 2016 10:07:27 +0000 (11:07 +0100)]
mem: Use pkt::getBlockAddr instead of BaseCace::blockAlign

Change-Id: I0ed4e528cb750a323facdc811dde7f0ed1ff228e
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
7 years agoarm, kvm: enable running 32-bit Guest under ARM KVM64
Rahul Thakur [Wed, 1 Mar 2017 18:27:05 +0000 (10:27 -0800)]
arm, kvm: enable running 32-bit Guest under ARM KVM64

1) Pass KVM_ARM_VCPU_EL1_32BIT to kvmArmVCpuInit
   when running 32-bit OS

2) Correctly map 64-bit registers to banked 32-bit ones

Change-Id: I1dec6427d6f5c3bba599ccdd804f1dfe80d3e670
Reviewed-on: https://gem5-review.googlesource.com/2261
Maintainer: Rahul Thakur <rjthakur@google.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
7 years agoarm, kvm: fix saving/restoring conditional flags in ARM KVM64
Rahul Thakur [Wed, 1 Mar 2017 18:15:57 +0000 (10:15 -0800)]
arm, kvm: fix saving/restoring conditional flags in ARM KVM64

The gem5 stores flags separately from other fields CPSR, so we need to
split them out and recombine on trips to/from KVM.

Change-Id: I28ed00eb6f0e2a1436adfbc51b6ccf056958afeb
Reviewed-on: https://gem5-review.googlesource.com/2260
Reviewed-by: Rahul Thakur <rjthakur@google.com>
Maintainer: Rahul Thakur <rjthakur@google.com>

7 years agoconfig: exit with fatal() if error
Pierre-Yves Péneau [Thu, 23 Feb 2017 09:12:05 +0000 (10:12 +0100)]
config: exit with fatal() if error

If output redirection is activated, the error message is printed in
simout. This change ensure it will be printed in simerr.

Change-Id: Ie661ac6b6978bf2e4aaaccdf23134795d764d459
Signed-off-by: Pierre-Yves Péneau <pierre-yves.peneau@lirmm.fr>
Reviewed-on: https://gem5-review.googlesource.com/2221
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>

7 years agoruby: fix and/or precedence in slicc
Lena Olson [Sun, 5 Feb 2017 20:47:37 +0000 (14:47 -0600)]
ruby: fix and/or precedence in slicc

The slicc compiler currently treats && and || with the same precedence.
This is highly non-intuitive to people used to C, and was probably an
error. This patch makes && bind tighter than ||.

For example, previously:
if (A || B && C)
compiled to:
if ((A || B) && C)
With this patch, it compiles to:
if (A || (B && C))

Change-Id: Idbbd5b50cc86a8d6601045adc14a253284d7b791
Signed-off-by: Lena Olson (leolson@google.com)
Reviewed-on: https://gem5-review.googlesource.com/2168
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Joe Gross <criusx@gmail.com>
Reviewed-by: Sooraj Puthoor <puthoorsooraj@gmail.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
[ Rebased onto master ]
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
7 years agoscons: Automatically add a git commit message hook
Andreas Sandberg [Sun, 5 Feb 2017 05:00:38 +0000 (05:00 +0000)]
scons: Automatically add a git commit message hook

Gerrit requires that all commit messages have a Change-Id tag. This
tag is added automatically by a commit message hook in Git. Include
the default Gerrit commit message hook and add it automatically using
scons to make life easier for everyone.

Change-Id: I1270fbaaadf6ed151bddf14521a38e0c1a02d131
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2166
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>

7 years agosyscall_emul: [PATCH 15/22] add clone/execve for threading and multiprocess simulations
Brandon Potter [Mon, 27 Feb 2017 19:10:15 +0000 (14:10 -0500)]
syscall_emul: [PATCH 15/22] add clone/execve for threading and multiprocess simulations

Modifies the clone system call and adds execve system call. Requires allowing
processes to steal thread contexts from other processes in the same system
object and the ability to detach pieces of process state (such as MemState)
to allow dynamic sharing.

7 years agosyscall_emul: [patch 14/22] adds identifier system calls
Brandon Potter [Mon, 27 Feb 2017 19:10:02 +0000 (14:10 -0500)]
syscall_emul: [patch 14/22] adds identifier system calls

This changeset add fields to the process object and adds the following
three system calls: setpgid, gettid, getpid.

7 years agox86: remove unnecessary parameter from functions
Brandon Potter [Mon, 27 Feb 2017 19:09:30 +0000 (14:09 -0500)]
x86: remove unnecessary parameter from functions

7 years agogpu-compute: remove unnecessary member from class
Tony Gutierrez [Mon, 27 Feb 2017 18:18:51 +0000 (13:18 -0500)]
gpu-compute: remove unnecessary member from class

The clang compiler complains that the wavefront member in
the GpuISA class is unused. This changeset removes the member,
because it does not appear serve a purpose.

7 years agogpu-compute: mark functions with override if replacing virtual
Brandon Potter [Mon, 27 Feb 2017 18:18:38 +0000 (13:18 -0500)]
gpu-compute: mark functions with override if replacing virtual

The clang compiler is more stringent than the recent versions of
GCC when dealing with overrides. This changeset adds the specifier
to the methods which need it to silence the compiler.

7 years agoarch: Include generated decoder header after normal headers
Andreas Sandberg [Mon, 27 Feb 2017 12:06:00 +0000 (12:06 +0000)]
arch: Include generated decoder header after normal headers

The generated decoder header defines macros that represent bit fields
within instructions. These fields typically have short names that
conflict with names in other header files. Include the generated
header after all normal header to avoid this issue.

Change-Id: I53d149b75432c20abdbf651e32c3c785d897973b
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
7 years agobase: Refactor logging to make log level selection cleaner
Andreas Sandberg [Mon, 27 Feb 2017 11:25:01 +0000 (11:25 +0000)]
base: Refactor logging to make log level selection cleaner

It's currently possible to change the log level in gem5 by tweaking a
set of global variables. These variables are currently exposed to
Python using SWIG. This mechanism is far from ideal for two reasons:
First, changing the log level requires that the Python world enables
or disables individual levels. Ideally, this should be a single call
where a log level is selected. Second, exporting global variables is
poorly supported by most Python frameworks. SWIG puts variables in
their own namespace and PyBind doesn't seem to support it at all.

This changeset refactors the logging code to create a more abstract
interface. Each log level is associated with an instance of a Logger
class. This class contains common functionality, an enable flag, and a
verbose flag.

Available LogLevels are described by the LogLevel class. Lower log
levels are used for more critical messages (PANIC being level 0) and
higher levels for less critical messages. The highest log level that
is printed is controlled by calling Logger:setLevel().

Change-Id: I31e44299d242d953197a8e62679250c91d6ef776
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Gabor Dozsa <gabor.dozsa@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
7 years agotests: Disable descriptions in stat files
Andreas Sandberg [Mon, 27 Feb 2017 11:25:00 +0000 (11:25 +0000)]
tests: Disable descriptions in stat files

Don't output verbose text descriptions in stat files when running
tests. This saves a lot of space when storing reference data.

Change-Id: I2a7ead4843586e800ecf83846694b73f0c356373
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Sascha Bischoff <sascha.bischoff@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Tony Gutierrez <anthony.gutierrez@amd.com>
7 years agopython: Add a generalized mechanism to configure stats
Andreas Sandberg [Mon, 27 Feb 2017 11:24:59 +0000 (11:24 +0000)]
python: Add a generalized mechanism to configure stats

Add a mechanism to configure the stat output format using a URL-like
syntax. This makes it possible to specify both an output format
(currently, only text is supported) and override default
parameters.

On the Python-side, this is implemented using a helper function
(m5.stats.addStatVisitor) that adds a visitor to the list of active
stat visitors. The helper function parses a URL-like stat
specification to determine the stat output type. Optional parameters
can be specified to change how stat visitors behave.

For example, to output stats in text format without stat descriptions:

    m5.stats.addStatVisitor("text://stats.txt?desc=False")

From the command line:

    gem5.opt --stats-file="text://stats.txt?desc=False"

Internally, the stat framework uses the _url_factory decorator
to wrap a Python function with the fn(path, **kwargs) signature in a
function that takes a parsed URL as its only argument. The path and
keyword arguments are automatically derived from the URL in the
wrapper function.

New output formats can be registered in the m5.stats.factories
dictionary. This dictionary contains a mapping between format names
(URL schemes) and factory methods.

To retain backwards compatibility, the code automatically assumes that
the user wants text output if no format has been specified (i.e., when
specifying a plain path).

Change-Id: Ic4dce93ab4ead07ffdf71e55a22ba0ae5a143061
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: Ilias Vougioukas <ilias.vougioukas@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Tony Gutierrez <anthony.gutierrez@amd.com>
9 years agosyscall_emul: [patch 13/22] add system call retry capability
Brandon Potter [Mon, 20 Jul 2015 14:15:21 +0000 (09:15 -0500)]
syscall_emul: [patch 13/22] add system call retry capability

This changeset adds functionality that allows system calls to retry without
affecting thread context state such as the program counter or register values
for the associated thread context (when system calls return with a retry
fault).

This functionality is needed to solve problems with blocking system calls
in multi-process or multi-threaded simulations where information is passed
between processes/threads. Blocking system calls can cause deadlock because
the simulator itself is single threaded. There is only a single thread
servicing the event queue which can cause deadlock if the thread hits a
blocking system call instruction.

To illustrate the problem, consider two processes using the producer/consumer
sharing model. The processes can use file descriptors and the read and write
calls to pass information to one another. If the consumer calls the blocking
read system call before the producer has produced anything, the call will
block the event queue (while executing the system call instruction) and
deadlock the simulation.

The solution implemented in this changeset is to recognize that the system
calls will block and then generate a special retry fault. The fault will
be sent back up through the function call chain until it is exposed to the
cpu model's pipeline where the fault becomes visible. The fault will trigger
the cpu model to replay the instruction at a future tick where the call has
a chance to succeed without actually going into a blocking state.

In subsequent patches, we recognize that a syscall will block by calling a
non-blocking poll (from inside the system call implementation) and checking
for events. When events show up during the poll, it signifies that the call
would not have blocked and the syscall is allowed to proceed (calling an
underlying host system call if necessary). If no events are returned from the
poll, we generate the fault and try the instruction for the thread context
at a distant tick. Note that retrying every tick is not efficient.

As an aside, the simulator has some multi-threading support for the event
queue, but it is not used by default and needs work. Even if the event queue
was completely multi-threaded, meaning that there is a hardware thread on
the host servicing a single simulator thread contexts with a 1:1 mapping
between them, it's still possible to run into deadlock due to the event queue
barriers on quantum boundaries. The solution of replaying at a later tick
is the simplest solution and solves the problem generally.

9 years agostyle: [patch 12/22] fix preliminary style issues for subsequent fault patch
Brandon Potter [Mon, 20 Jul 2015 14:15:21 +0000 (09:15 -0500)]
style: [patch 12/22] fix preliminary style issues for subsequent fault patch

This changeset add spaces in a few spots and removes an unnecessary comment.

9 years agosyscall_emul: [patch 11/22] extend functionality of fcntl
Brandon Potter [Mon, 20 Jul 2015 14:15:21 +0000 (09:15 -0500)]
syscall_emul: [patch 11/22] extend functionality of fcntl

This changeset adds the ability to set a close-on-exec flag for a given
file descriptor. It also reworks some of the logic surrounding setting and
retrieving flags from the file description.

7 years agox86: remove redundant condition check in tlb code
Brandon Potter [Thu, 23 Feb 2017 18:27:48 +0000 (13:27 -0500)]
x86: remove redundant condition check in tlb code

7 years agobase: fix small memory leak in the ELF loader
Brandon Potter [Thu, 23 Feb 2017 18:27:38 +0000 (13:27 -0500)]
base: fix small memory leak in the ELF loader

7 years agomem: Remove unused size field from the CacheBlk class
Nikos Nikoleris [Tue, 21 Feb 2017 14:14:45 +0000 (14:14 +0000)]
mem: Remove unused size field from the CacheBlk class

Change-Id: I6149290d6d2ac1a4bd6165871c93d7b7d6a980ad
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
7 years agomem: Remove the unused asid field from the CacheBlk class
Nikos Nikoleris [Tue, 21 Feb 2017 14:14:45 +0000 (14:14 +0000)]
mem: Remove the unused asid field from the CacheBlk class

Change-Id: I29f45733c5fad822bdd0d8dcc7939d86b2e8c97b
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
7 years agomem: Remove unused arguments (asid/contex_id) from accessBlock
Nikos Nikoleris [Tue, 21 Feb 2017 14:14:44 +0000 (14:14 +0000)]
mem: Remove unused arguments (asid/contex_id) from accessBlock

Change-Id: I79c2662fc81630ab321db8a75be6cd15fa07d372
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
7 years agomem: Remove unused type BlkList from the cache and the tags
Nikos Nikoleris [Tue, 21 Feb 2017 14:14:44 +0000 (14:14 +0000)]
mem: Remove unused type BlkList from the cache and the tags

Change-Id: If9ebb8488e8db587482ecfa99d2c12cfe5734fb9
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
7 years agomem: Remove unused functions from the tag classes
Nikos Nikoleris [Tue, 21 Feb 2017 14:14:44 +0000 (14:14 +0000)]
mem: Remove unused functions from the tag classes

Change-Id: I4f3c2c027b1acaaf791a4c71086f34a9b9fbf4df
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
7 years agomem: Always use the helper function to invalidate a block
Nikos Nikoleris [Tue, 21 Feb 2017 14:14:44 +0000 (14:14 +0000)]
mem: Always use the helper function to invalidate a block

Policies like the LRU need to be notified when a block is invalidated,
the helper function does this along with invalidating the block.

Change-Id: I3ed59cf07938caa7f394ee6054b0af9e00b267ea
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
7 years agomem: Fix MSHR assert triggering for invalidated prefetches
Sascha Bischoff [Tue, 21 Feb 2017 14:14:44 +0000 (14:14 +0000)]
mem: Fix MSHR assert triggering for invalidated prefetches

This changeset updates an assert in src/mem/cache/mshr.cc which was
erroneously catching invalidated prefetch requests. These requests can
become invalidated if another component writes (an exclusive access)
to this location during the time that the read request is in
flight. The original assert made the assumption that these cases can
only occur for reads generated by the CPU, and hence
prefetcher-generated requests would sometimes trip the assert.

Change-Id: If4f043273a688c2bab8f7a641192a2b583e7b20e
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
7 years agomem: Populate the secure flag in the writeback visitor
Nikos Nikoleris [Tue, 21 Feb 2017 14:14:44 +0000 (14:14 +0000)]
mem: Populate the secure flag in the writeback visitor

Previously the writeback visitor would not consider and set the secure
flag for the blocks that are written back to memory. This patch fixes
this.

Change-Id: Ie1a425fa9211407a70a4343f2c6b3d073371378f
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
7 years agomisc: Add dtb files to the ignore list for git and mercurial
Nikos Nikoleris [Tue, 21 Feb 2017 14:14:44 +0000 (14:14 +0000)]
misc: Add dtb files to the ignore list for git and mercurial

Change-Id: Ifb135c60e050c55769914e853b07a387c06e4007
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
7 years agomem: Remove stale argument from a panic statement
Nikos Nikoleris [Tue, 21 Feb 2017 14:14:44 +0000 (14:14 +0000)]
mem: Remove stale argument from a panic statement

Change-Id: I7ae5fa44a937f641a2ddd242a49e0cd23f68b9f2
Reviewed-by: Sudhanshu Jha <sudhanshu.jha@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
7 years agoarm: Fix DPRINTFs with arguments in the instruction declarations
Nikos Nikoleris [Tue, 21 Feb 2017 14:14:44 +0000 (14:14 +0000)]
arm: Fix DPRINTFs with arguments in the instruction declarations

Change-Id: I0e373536897aa5bb4501b00945c2a0836100ddf4
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
7 years agoarm: Blame the right instruction address on a Prefetch Abort
Nikos Nikoleris [Tue, 21 Feb 2017 14:14:44 +0000 (14:14 +0000)]
arm: Blame the right instruction address on a Prefetch Abort

CPU models (e.g., O3CPU) issue instruction fetches for the whole cache
block rather than a specific instruction. Consequently the TLB lookups
translate the cache block virtual address. When the TLB lookup fails,
however, the Prefetch Abort must be raised for the PC of the
instruction that caused the fault rather than for the address of the
block.

This change fixes the way we instantiate the PrefetchAbort faults to
use the PC of the request rather the address of the instruction fetch
request.

Change-Id: I8e45549da1c3be55ad204a060029c95ce822a851
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-by: Rekai Gonzalez Alberquilla <rekai.gonzalezalberquilla@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
7 years agostats: Get all stats updated to reflect current behaviour
Andreas Hansson [Sun, 19 Feb 2017 10:30:32 +0000 (05:30 -0500)]
stats: Get all stats updated to reflect current behaviour

Line everything up again.

7 years agosim: Ensure draining is deterministic
Andreas Hansson [Sun, 19 Feb 2017 10:30:31 +0000 (05:30 -0500)]
sim: Ensure draining is deterministic

The traversal of drainable objects could potentially be
non-deterministic when using an unordered set containing object
pointers. To ensure that the iteration is deterministic, we switch to
a vector. Note that the lookup and traversal of the drainable objects
is not performance critical, so the change has no negative consequences.

7 years agomem: Ensure deferred snoops are cache-line aligned
Andreas Hansson [Sun, 19 Feb 2017 10:30:31 +0000 (05:30 -0500)]
mem: Ensure deferred snoops are cache-line aligned

This patch fixes a bug where a deferred snoop ended up being to a
partial cache line, and not cache-line aligned, all due to how we copy
the packet.

7 years agomem: Fix memory footprint includes
Andreas Hansson [Sun, 19 Feb 2017 10:30:31 +0000 (05:30 -0500)]
mem: Fix memory footprint includes

Fix compilation errors due to missing include.

8 years agosyscall_emul: [patch 10/22] refactor fdentry and add fdarray class
Brandon Potter [Wed, 9 Nov 2016 20:27:42 +0000 (14:27 -0600)]
syscall_emul: [patch 10/22] refactor fdentry and add fdarray class

Several large changes happen in this patch.

The FDEntry class is rewritten so that file descriptors now correspond to
types: 'File' which is normal file-backed file with the file open on the
host machine, 'Pipe' which is a pipe that has been opened on the host machine,
and 'Device' which does not have an open file on the host yet acts as a pseudo
device with which to issue ioctls. Other types which might be added in the
future are directory entries and sockets (off the top of my head).

The FDArray class was create to hold most of the file descriptor handling
that was stuffed into the Process class. It uses shared pointers and
the std::array type to hold the FDEntries mentioned above.

The changes to these two classes needed to be propagated out to the rest
of the code so there were quite a few changes for that. Also, comments were
added where I thought they were needed to help others and extend our
DOxygen coverage.

8 years agosyscall_emul: [patch 9/22] remove unused global variable (num_processes)
Brandon Potter [Wed, 9 Nov 2016 20:27:42 +0000 (14:27 -0600)]
syscall_emul: [patch 9/22] remove unused global variable (num_processes)

8 years agosyscall_emul: [patch 8/22] refactor process class
Brandon Potter [Wed, 9 Nov 2016 20:27:41 +0000 (14:27 -0600)]
syscall_emul: [patch 8/22] refactor process class

Moves aux_vector into its own .hh and .cc files just to get it out of the
already crowded Process files. Arguably, it could stay there, but it's
probably better just to move it and give it files.

The changeset looks ugly around the Process header file, but the goal here is
to move methods and members around so that they're not defined randomly
throughout the entire header file. I expect this is likely one of the reasons
why I several unused variables related to this class. So, the methods are
declared first followed by members. I've tried to aggregate them together
so that similar entries reside near one another.

There are other changes coming to this code so this is by no means the
final product.

8 years agosyscall_emul: [patch 7/22] remove numCpus method
Brandon Potter [Wed, 9 Nov 2016 20:27:41 +0000 (14:27 -0600)]
syscall_emul: [patch 7/22] remove numCpus method

The numCpus method is misleading in that it's not really a measure of
how many CPUs might be executing a process, but how many thread contexts
are assigned to the process at any given point in time.

It's nice to highlight this distinction because thread contexts are never
reused in the same way that a CPU can be reused for multiple processes.
The reason that there is no reuse is that there is no CPU scheduler for SE.

The tru64 code intends to use this method and the accompanying contextIDs
field to support SMT and track the number of threads with some system calls.
With the up coming clone and exec patches, this paradigm must change. There
needs to be a 1:1 mapping between the thread contexts and processes so that
the process state between threads is allowed to vary when needed by Linux.
This should not break SMT for tru64 if the Process class is refactored so that
multiple Processes can share state between themselves. The following patches
will do the refactoring incrementally as features are added.

8 years agosyscall_emul: [patch 6/22] remove unused fields from Process class
Brandon Potter [Wed, 9 Nov 2016 20:27:41 +0000 (14:27 -0600)]
syscall_emul: [patch 6/22] remove unused fields from Process class

It looks like tru64 has some nxm* system calls, but the two fields that
are defined in the Process class are unused by any of the code. There doesn't
appear to be any reference in the tru64 code.

8 years agosyscall_emul: [patch 5/22] remove LiveProcess class and use Process instead
Brandon Potter [Wed, 9 Nov 2016 20:27:40 +0000 (14:27 -0600)]
syscall_emul: [patch 5/22] remove LiveProcess class and use Process instead

The EIOProcess class was removed recently and it was the only other class
which derived from Process. Since every Process invocation is also a
LiveProcess invocation, it makes sense to simplify the organization by
combining the fields from LiveProcess into Process.

7 years agosparc: fix bugs caused by cd7f3a1dbf55
Brandon Potter [Fri, 17 Feb 2017 17:01:51 +0000 (12:01 -0500)]
sparc: fix bugs caused by cd7f3a1dbf55

Turns out that SPARC SE mode relied on M5_pid being "0" in
all cases. The entries in the SPARC TLBs are accessed with
M5_pid as their context. This is buggy in the sense that it
will never work with more than one process or any
initialization that doesn't have the M5_pid value passed in
as "0".

cd7f3a1dbf55 broke the SPARC build because it deletes M5_pid
and uses a _pid with a default of "100" instead. This caused
the SPARC TLB to never return any valid lookups for any
request; the program never moved past the first instruction
with SPARC SE in the regression tester.

The solution proposed in this changeset is to initialize
the address space identification register with the PID value
that is passed into the process class as a parameter from
Python. This should return the correct responses from the TLB
since the insertions and lookups into the page table will be
using the same PID.

Furthermore, there are corner cases in the code which elevate
privileges and revert to using context "0" as the context in
the TLB. I believe that these are related to kernel level
traps and hypervisor privilege escalations, but I'm not
completely sure. I've tried to address the corner cases
properly, but it would be beneficial to have someone who is
familiar with the SPARC architecture to take a look at this
fix.

7 years agosim: fix out-of-bounds error in syscall_desc
Brandon Potter [Fri, 17 Feb 2017 17:01:50 +0000 (12:01 -0500)]
sim: fix out-of-bounds error in syscall_desc

7 years agomem, stats: fix typos in CommMonitor and Stats
Pierre-Yves Péneau [Wed, 15 Feb 2017 20:59:06 +0000 (14:59 -0600)]
mem, stats: fix typos in CommMonitor and Stats

Signed-off-by: Pierre-Yves Péneau <pierre-yves.peneau@lirmm.fr>
Reviewed-by: Tony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Signed-off-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed at http://reviews.gem5.org/r/3802/

7 years agomem, misc: fix building issue with CommMonitor (unused variables)
Pierre-Yves Péneau [Wed, 15 Feb 2017 20:56:54 +0000 (14:56 -0600)]
mem, misc: fix building issue with CommMonitor (unused variables)

Signed-off-by: Pierre-Yves Péneau <pierre-yves.peneau@lirmm.fr>
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Signed-off-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed at http://reviews.gem5.org/r/3801/