gem5.git
8 years agogpu-compute: move disassemle() implementation to GPUStaticInst
Tony Gutierrez [Thu, 27 Oct 2016 02:47:05 +0000 (22:47 -0400)]
gpu-compute: move disassemle() implementation to GPUStaticInst

8 years agogpu-compute, arch: add some methods to the base inst classes for ISA support
Tony Gutierrez [Thu, 27 Oct 2016 02:47:01 +0000 (22:47 -0400)]
gpu-compute, arch: add some methods to the base inst classes for ISA support

8 years agoruby: make a RequestDesc class instead of std::pair
Tony Gutierrez [Thu, 27 Oct 2016 02:46:58 +0000 (22:46 -0400)]
ruby: make a RequestDesc class instead of std::pair

the RequestDesc was previously implemented as a std::pair, which made
the implementation overly complex and error prone. here we encapsulate the
packet, primary, and secondary types all in a single data structure with
all members properly intialized in a ctor

8 years agoconfig: Break out base options for usage with NULL ISA
Andreas Hansson [Wed, 26 Oct 2016 18:50:54 +0000 (14:50 -0400)]
config: Break out base options for usage with NULL ISA

This patch breaks out the most basic configuration options into a set
of base options, to allow them to be used also by scripts that do not
involve any ISA, and thus no actual CPUs or devices.

The patch also fixes a few modules so that they can be imported in a
NULL build, and avoid dragging in FSConfig every time Options is
imported.

8 years agostats: Update stats to reflect recent changes to floats
Andreas Hansson [Wed, 19 Oct 2016 10:20:04 +0000 (06:20 -0400)]
stats: Update stats to reflect recent changes to floats

Mostly just splitting out the floats ops and corresponding
reads/writes.

8 years agoarm: Fix for ARM's Streamline conversion script
Shawn Rosti [Sat, 15 Oct 2016 20:11:07 +0000 (15:11 -0500)]
arm: Fix for ARM's Streamline conversion script

tracked down issue with ARM's version of gem5 using the "cluster" name.
The public/github version of ARM Gem5 does not use the "cluster" naming
mechanism.

Signed-off-by: Dam Sunwoo <dam.sunwoo@arm.com>
Signed-off-by: Jason Lowe-Power <jason@lowepower.com>
8 years agoarm, dev: pl011 console interactivity
Bjoern A. Zeeb [Sat, 15 Oct 2016 20:11:04 +0000 (15:11 -0500)]
arm, dev: pl011 console interactivity

Improve PL011 console interactivity

Signed-off-by: Jason Lowe-Power <jason@lowepower.com>
8 years agosyscall: read() should not write anything if reading EOF.
Nicolas Derumigny [Sat, 15 Oct 2016 20:06:24 +0000 (15:06 -0500)]
syscall: read() should not write anything if reading EOF.

Read() should not write anything when returning 0 (EOF).
This patch does not correct the same bug occuring for :

nbr_read=read(file, buf, nbytes)

When nbr_read<nbytes, nbytes bytes are copied into the virtual
RAM instead of nbr_read. If buf is smaller than nbytes, a
page fault occurs, even if buf is in fact bigger than nbr_read.

Signed-off-by: Jason Lowe-Power <jason@lowepower.com>
8 years agocpu, arm: Distinguish Float* and SimdFloat*, create FloatMem* opClass
Fernando Endo [Sat, 15 Oct 2016 19:58:45 +0000 (14:58 -0500)]
cpu, arm: Distinguish Float* and SimdFloat*, create FloatMem* opClass

Modify the opClass assigned to AArch64 FP instructions from SimdFloat* to
Float*. Also create the FloatMemRead and FloatMemWrite opClasses, which
distinguishes writes to the INT and FP register banks.
Change the latency of (Simd)FloatMultAcc to 5, based on the Cortex-A72,
where the "latency" of FMADD is 3 if the next instruction is a FMADD and
has only the augend to destination dependency, otherwise it's 7 cycles.

Signed-off-by: Jason Lowe-Power <jason@lowepower.com>
8 years agoconfig: Make configs/common a Python package
Andreas Hansson [Fri, 14 Oct 2016 14:37:38 +0000 (10:37 -0400)]
config: Make configs/common a Python package

Continue along the same line as the recent patch that made the
Ruby-related config scripts Python packages and make also the
configs/common directory a package.

All affected config scripts are updated (hopefully).

Note that this change makes it apparent that the current organisation
and naming of the config directory and its subdirectories is rather
chaotic. We mix scripts that are directly invoked with scripts that
merely contain convenience functions. While it is not addressed in
this patch we should follow up with a re-organisation of the
config structure, and renaming of some of the packages.

8 years agostats: Add more information to uninitialized error
Jason Lowe-Power [Fri, 14 Oct 2016 14:02:03 +0000 (09:02 -0500)]
stats: Add more information to uninitialized error

ClockedObject was changed to require its regStats() to be called from every
child class. If you forget to do this, the error was indecipherable. This
patch makes the error more clear.

8 years agostats: update references
Curtis Dunham [Thu, 13 Oct 2016 22:21:40 +0000 (23:21 +0100)]
stats: update references

8 years agomem: add DRAM powerdown current
Omar Naji [Thu, 13 Oct 2016 18:22:11 +0000 (19:22 +0100)]
mem: add DRAM powerdown current

Change-Id: I763cffe0c69f5ebbbf6a6eb12bec5c13d5d0161d
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
Reviewed-by: Radhika Jagtap <radhika.jagtap@arm.com>
8 years agomem: Add DRAM low-power functionality
Wendy Elsasser [Thu, 13 Oct 2016 18:22:11 +0000 (19:22 +0100)]
mem: Add DRAM low-power functionality

Added power-down state transitions to the DRAM controller model.

Added per rank parameter, outstandingEvents, which tracks the number
of outstanding command events and is used to determine when the
controller should transition to a low power state.
The controller will only transition when there are no outstanding events
scheduled and the number of command entries for the given rank is 0.

The outstandingEvents parameter is incremented for every RD/WR burst,
PRE, and REF event scheduled.  ACT is implicitly covered by RD/WR
since burst will always issue and complete after a required ACT.
The parameter is decremented when the event is serviced (completed).

The controller will automatically transition to ACT power down,
PRE power down, or SREF.

Transition to ACT power down state scheduled from:
1) The RespondEvent, where read data is received from the memory.
   ACT power-down entry will be scheduled when one or more banks is
   open, all commands for the rank have completed (no more commands
   scheduled), and there are no commands in queue for the rank

Transition to PRE power down scheduled from:
1) respondEvent, when all banks are closed, all commands have
   completed, and there are no commands in queue for the rank
2) prechargeEvent when all banks are closed, all commands have
   completed, and there are no commands in queue for the rank
3) refreshEvent, after the refresh is complete when the previous
   state was ACT power-down
4) refreshEvent, after the refresh is complete when the previous
   state was PRE power-down and there are commands in the queue.

Transition to SREF will be scheduled from:
1) refreshEvent, after the refresh is completes when the previous
   state was PRE power-down with no commands in queue

Power-down exit commands are scheduled from:
1) The refreshEvent, prior to issuing a refresh
2) doDRAMAccess, to wake-up the rank for RD/WR command issue.

Self-refresh exit commands are scheduled from:
1) The next request event, when the queue has commands for the rank
   in the readQueue or there are commands for the rank in the
   writeQueue and the bus state is WRITE.

Change-Id: I6103f660776e36c686655e71d92ec7b5b752050a
Reviewed-by: Radhika Jagtap <radhika.jagtap@arm.com>
8 years agomem: Add callback to compute stats prior to dump event
Wendy Elsasser [Thu, 13 Oct 2016 18:22:11 +0000 (19:22 +0100)]
mem: Add callback to compute stats prior to dump event

The per rank statistics are periodically updated based on
state transition and refresh events.

Add a method to update these when a dump event occurs to
ensure they reflect accurate values.
Specifically, need to ensure that the low-power state
durations, power, and energy are logged correctly.

Change-Id: Ib642a6668340de8f494a608bb34982e58ba7f1eb
Reviewed-by: Radhika Jagtap <radhika.jagtap@arm.com>
8 years agomem: Modify drain to ensure banks and power are idled
Wendy Elsasser [Thu, 13 Oct 2016 18:22:11 +0000 (19:22 +0100)]
mem: Modify drain to ensure banks and power are idled

Add constraint that all ranks have to be in PWR_IDLE
before signaling drain complete

This will ensure that the banks are all closed and the rank
has exited any low-power states.

On suspend, update the power stats to sync the DRAM power logic

The logic maintains the location of the signalDrainDone
method, which is still triggered from either:
1) Read response event
2) Next request event

This ensures that the drain will complete in the READ bus
state and minimizes the changes required.

Change-Id: If1476e631ea7d5999fe50a0c9379c5967a90e3d1
Reviewed-by: Radhika Jagtap <radhika.jagtap@arm.com>
8 years agomem: Sort memory commands and update DRAMPower
Wendy Elsasser [Thu, 13 Oct 2016 18:22:10 +0000 (19:22 +0100)]
mem: Sort memory commands and update DRAMPower

Add local variable to stores commands to be issued.
These commands are in order within a single bank but will be out
of order across banks & ranks.

A new procedure, flushCmdList, sorts commands across banks / ranks,
and flushes the sorted list, up to curTick() to DRAMPower.
This is currently called in refresh, once all previous commands are
guaranteed to have completed.  Could be called in other events like
the powerEvent as well.

By only flushing commands up to curTick(), will not get out of sync
when flushed at a periodic stats dump (done in subsequent patch).

Change-Id: I4ac65a52407f64270db1e16a1fb04cfe7f638851
Reviewed-by: Radhika Jagtap <radhika.jagtap@arm.com>
8 years agomem: update DDR3 die revision
Omar Naji [Thu, 13 Oct 2016 18:22:10 +0000 (19:22 +0100)]
mem: update DDR3 die revision

Change-Id: I8992ddc1664c3ed4b2d36d8a34e4ce8be113b9de
Reviewed-by: Radhika Jagtap <radhika.jagtap@arm.com>
8 years agomem: add DRAM powerdown timing
Omar Naji [Thu, 13 Oct 2016 18:22:10 +0000 (19:22 +0100)]
mem: add DRAM powerdown timing

8 years agomem: make DDR4 x16
Omar Naji [Thu, 13 Oct 2016 18:22:10 +0000 (19:22 +0100)]
mem: make DDR4 x16

8 years agoisa,arm: Add missing AArch32 FP instructions
Mitch Hayenga [Thu, 13 Oct 2016 18:22:10 +0000 (19:22 +0100)]
isa,arm: Add missing AArch32 FP instructions

This commit adds missing non-predicated, scalar floating point
instructions.  Specifically VRINT* floating point integer rounding
instructions and VSEL* floating point conditional selects.

Change-Id: I23cbd1389f151389ac8beb28a7d18d5f93d000e7
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Nathanael Premillieu <nathanael.premillieu@arm.com>
8 years agoruby: Fix regressions and make Ruby configs Python packages
Andreas Hansson [Thu, 13 Oct 2016 07:17:19 +0000 (03:17 -0400)]
ruby: Fix regressions and make Ruby configs Python packages

This patch moves the addition of network options into the Ruby module
to avoid the regressions all having to add it explicitly. Doing this
exposes an issue in our current config system though, namely the fact
that addtoPath is relative to the Python script being executed. Since
both example and regression scripts use the Ruby module we would end
up with two different (relative) paths being added. Instead we take a
first step at turning the config modules into Python packages, simply
by adding a __init__.py in the configs/ruby, configs/topologies and
configs/network subdirectories.

As a result, we can now add the top-level configs directory to the
Python search path, and then use the package names in the various
modules. The example scripts are also updated, and the messy
path-deducing variations in the scripts are unified.

8 years agoconfig: fix typo in cluster topology.
Tushar Krishna [Sat, 8 Oct 2016 03:56:48 +0000 (23:56 -0400)]
config: fix typo in cluster topology.

8 years agodev, arm: Make GenericTimer param handling more robust
Andreas Sandberg [Fri, 7 Oct 2016 13:14:44 +0000 (14:14 +0100)]
dev, arm: Make GenericTimer param handling more robust

The generic timer needs a pointer to an ArmSystem to wire itself to the
system register handler. This was previously specified as an instance
of System that was later cast to ArmSystem. Make this more robust by
specifying it as an ArmSystem in the Python interface and add a check
to make sure that it is non-NULL.

Change-Id: I989455e666f4ea324df28124edbbadfd094b0d02
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
8 years agoruby: Add M5_VAR_USED before variables used only inside assert in garnet2.0.
Tushar Krishna [Fri, 7 Oct 2016 01:06:00 +0000 (21:06 -0400)]
ruby: Add M5_VAR_USED before variables used only inside assert in garnet2.0.
This removes errors when building gem5.fast

8 years agoruby: garnet2.0
Tushar Krishna [Thu, 6 Oct 2016 18:35:22 +0000 (14:35 -0400)]
ruby: garnet2.0
Revamped version of garnet with more optimized single-cycle routers,
more configurability, and cleaner code.

8 years agoruby: remove the original garnet code.
Tushar Krishna [Thu, 6 Oct 2016 18:35:21 +0000 (14:35 -0400)]
ruby: remove the original garnet code.
Only garnet2.0 will be supported henceforth.

8 years agoconfig: add port directions and per-router delay in topology.
Tushar Krishna [Thu, 6 Oct 2016 18:35:20 +0000 (14:35 -0400)]
config: add port directions and per-router delay in topology.
This patch adds port direction names to the links during topology
creation, which can be used for better printed names for the links
or for users to code up their own adaptive routing algorithms.
It also adds support for every router to have an independent latency
value to support heterogeneous topologies with the subsequent
garnet2.0 patch.

8 years agoconfig: make internal links in network topology unidirectional.
Tushar Krishna [Thu, 6 Oct 2016 18:35:18 +0000 (14:35 -0400)]
config: make internal links in network topology unidirectional.
This patch makes the internal links within the network topology
unidirectional, thus allowing any deadlock-free routing algorithms to
be specified from the topology itself using weights.
This patch also renames Mesh.py and MeshDirCorners.py to
Mesh_XY.py and MeshDirCorners_XY.py (Mesh with XY routing).
It also adds a Mesh_westfirst.py and CrossbarGarnet.py topologies.

8 years agoconfig: add a separate config file for the network.
Tushar Krishna [Thu, 6 Oct 2016 18:35:17 +0000 (14:35 -0400)]
config: add a separate config file for the network.
This patch adds a new file configs/network/Network.py to setup the network,
instead of doing that within Ruby.py.

8 years agoruby: rename networktest to garnet_synthetic_traffic.
Tushar Krishna [Thu, 6 Oct 2016 18:35:16 +0000 (14:35 -0400)]
ruby: rename networktest to garnet_synthetic_traffic.
networktest is essentially a collection of synthetic traffic patterns
for the network. The protocol name and the tester having the same name
led to multiple python configuration files with the same name, adding
confusion. This patch renames networktest to garnet_synthetic_traffic,
and also adds more synthetic traffic patterns.

8 years agoruby: rename ALPHA_Network_test protocol to Garnet_standalone.
Tushar Krishna [Thu, 6 Oct 2016 18:35:14 +0000 (14:35 -0400)]
ruby: rename ALPHA_Network_test protocol to Garnet_standalone.
Over the past 6 years, we realized that the protocol is essentially used
to run the garnet network in a standalone manner, and feed standard synthetic
traffic patterns through it.

8 years agokvm: Adding details to kvm page fault in x86
Alexandru Dutu [Tue, 4 Oct 2016 17:06:05 +0000 (13:06 -0400)]
kvm: Adding details to kvm page fault in x86
Adding details, e.g. rip, rsp etc. to the kvm pagefault exit when in SE mode.

8 years agomisc: Adds a warning in case gdb is attached multiple times
Alexandru Dutu [Tue, 4 Oct 2016 17:04:19 +0000 (13:04 -0400)]
misc: Adds a warning in case gdb is attached multiple times
Instead of scheduling another event, this patch adds a warning in case gdb
is attached multiple times and the first attachement event has not been
processed yet.

8 years agogpu-compute: Added method to compute the actual workgroup size
Alexandru Dutu [Tue, 4 Oct 2016 17:03:52 +0000 (13:03 -0400)]
gpu-compute: Added method to compute the actual workgroup size
This patch adds a method to the Wavefront class to compute the actual workgroup
size. This can be different from the maximum workgroup size specified when
launching the kernel through the NDRange object. Current solution is still not
optimal, as we are computing these for each wavefront and the dispatcher also
needs to have this information and can't actually call
Wavefront::computeActuallWgSz before the wavefronts are being created. A long
term solution would be to have a Workgroup class that deals with all these
details.

8 years agoconfig: Fix lat_mem_rd example script
Andreas Hansson [Tue, 4 Oct 2016 14:44:52 +0000 (15:44 +0100)]
config: Fix lat_mem_rd example script

Adjust the traffic generator time-out so that the script works out of
the box

Change-Id: I6b3b6b11f98b094ae3acdbe09488c26e4aeb0ab4
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
8 years agosim: Add a checkpoint function to test for entries
Andreas Sandberg [Tue, 4 Oct 2016 10:22:16 +0000 (11:22 +0100)]
sim: Add a checkpoint function to test for entries

When loading a checkpoint, it's sometimes desirable to be able to test
whether an entry within a secion exists. This is currently done
automatically in the UNSERIALIZE_OPT_SCALAR macro, but it isn't
possible to do for arrays, containers, or enums. Instead of adding
even more macros, add a helper function (CheckpointIn::entryExists())
that tests for the presence of an entry.

Change-Id: I4b4646b03276b889fd3916efefff3bd552317dbc
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
8 years agoruby: correct size for partial memory writes
Brad Beckmann [Thu, 29 Sep 2016 05:06:52 +0000 (01:06 -0400)]
ruby: correct size for partial memory writes

Fixed AbstractController::queueMemoryWritePartial to specify the
correct size for partial memory writes.

8 years agomem: minor dprintf fix to abstract mem
Brad Beckmann [Thu, 29 Sep 2016 05:06:33 +0000 (01:06 -0400)]
mem: minor dprintf fix to abstract mem

print number of bytes written as a decimal number, not hex

8 years agoarm: disable GIC extensions
Curtis Dunham [Thu, 22 Sep 2016 13:46:37 +0000 (14:46 +0100)]
arm: disable GIC extensions

Change-Id: If19b9c593b48ded1ea848f2d3710d4369ec8a221
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
8 years agotests, arm: Reinstate accidentally removed switcheroo tests
Andreas Sandberg [Thu, 22 Sep 2016 11:10:47 +0000 (12:10 +0100)]
tests, arm: Reinstate accidentally removed switcheroo tests

Two of the switcheroo tests were accidentally removed due to
unexpected Mercurial behavior.

Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
8 years agocpu: Fix the O3 CPU Drain
Rekai Gonzalez-Alberquilla [Thu, 22 Sep 2016 09:49:10 +0000 (10:49 +0100)]
cpu: Fix the O3 CPU Drain

The drain did not wait until stages were ready again. Therefore, as a
result of messages in the TimeBuffer being drain, the state after the
drain was not consistent and asserts fired in some places when the
draining happened after a stage got blocked, but before the notification
arrived to the previous stages.

Change-Id: Ib50b3b40b7f745b62c1eba2931dec76860824c71
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
8 years agotest: Make the memtest and memcheck tests functional only
Andreas Sandberg [Thu, 22 Sep 2016 09:49:09 +0000 (10:49 +0100)]
test: Make the memtest and memcheck tests functional only

The memtest and memcheck are not designed to test timing. Make them
functional only to make ref diffs less noisy in the future.

Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
8 years agotests: Make remaining switcheroo tests functional only
Andreas Sandberg [Thu, 22 Sep 2016 09:49:08 +0000 (10:49 +0100)]
tests: Make remaining switcheroo tests functional only

The switcheroo tests only really serve to check functional
correctness. Checking for stat differences in them just increases the
size of reference diffs.

Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
8 years agoext: update SST test config
Curtis Dunham [Tue, 20 Sep 2016 14:51:24 +0000 (15:51 +0100)]
ext: update SST test config

Align configuration with new SST change [1] requiring units for
memHierarchy's backend.mem_size parameter.

[1] https://github.com/sstsimulator/sst-elements/commit/c901abb4e79644ff18f5222c94f5dae012772e1e

Change-Id: I19fa09bec8aa453dc52d154598a4ebb20ea304d8

8 years agogpu-compute: fix typo in GPUDispatcher
Tony Gutierrez [Fri, 16 Sep 2016 18:47:19 +0000 (14:47 -0400)]
gpu-compute: fix typo in GPUDispatcher

8 years agohsail: Fix disassembly of load instruction with 3 destination operands
Alexandru Dutu [Fri, 16 Sep 2016 16:36:20 +0000 (12:36 -0400)]
hsail: Fix disassembly of load instruction with 3 destination operands

8 years agogpu-compute: Adding context serialization methods to Wavefront
Alexandru Dutu [Fri, 16 Sep 2016 16:32:36 +0000 (12:32 -0400)]
gpu-compute: Adding context serialization methods to Wavefront
This patch adds methods to serialize the context of a particular wavefront
to the simulated system memory. Context serialization is used when a wavefront
is preempeted (i.e. context switch).

8 years agogpu-compute: Refactoring Wavefront::dynWaveId
Alexandru Dutu [Fri, 16 Sep 2016 16:31:46 +0000 (12:31 -0400)]
gpu-compute: Refactoring Wavefront::dynWaveId

8 years agogpu-compute: Adding vector register file debug messages
Alexandru Dutu [Fri, 16 Sep 2016 16:30:05 +0000 (12:30 -0400)]
gpu-compute: Adding vector register file debug messages
This patch introduces DPRINTFs for reading and writing to and from the vector
register file.

8 years agogpu-compute: Changing reconvergenceStack type
Alexandru Dutu [Fri, 16 Sep 2016 16:29:01 +0000 (12:29 -0400)]
gpu-compute: Changing reconvergenceStack type
std::stack has no iterators, therefore the reconvergence stack can't be
iterated without poping elements off. We will be using std::list instead to be
able to iterate for saving and restoring purposes.

8 years agogpu-compute: Adding ioctl for HW context size
Alexandru Dutu [Fri, 16 Sep 2016 16:27:56 +0000 (12:27 -0400)]
gpu-compute: Adding ioctl for HW context size
Adding runtime support for determining the memory required by a SIMD engine
when executing a particular wavefront.

8 years agogpu-compute: Wavefront refactoring
Alexandru Dutu [Fri, 16 Sep 2016 16:26:52 +0000 (12:26 -0400)]
gpu-compute: Wavefront refactoring
Renaming members of the Wavefront class in accordance with the style guide.

8 years agogpu-compute: Remove WFContext
Alexandru Dutu [Fri, 16 Sep 2016 16:26:03 +0000 (12:26 -0400)]
gpu-compute: Remove WFContext
WFContext struct is currently unused and it has been rendered not useful in
saving and restoring the context of a Wavefront. Wavefront class should be
sufficient for that purpose and the runtime can figure out the memory size
it will need to allocate for a Wavefront through an IOCTL.

8 years agotests, arm: Make switcheroo and checkpoint tests functional
Andreas Sandberg [Fri, 16 Sep 2016 08:14:31 +0000 (09:14 +0100)]
tests, arm: Make switcheroo and checkpoint tests functional

Switcheroo and checkpoint tests should generally be considered to be
successful if they run to completion.  Remove all reference output
files from the switcheroo and checkopint tests to make them purely
functional.

Change-Id: I70b47853bd662b7a33716d9e0d2154b16077f9dc
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
8 years agotests: Add support for functional only tests
Andreas Sandberg [Fri, 16 Sep 2016 08:04:20 +0000 (09:04 +0100)]
tests: Add support for functional only tests

Modify the ClassicTest class to only emit a stat verification test
unit if there is a reference stat file. This makes it possible to
design tests that don't care about stat changes.

To generate purely functional tests, we need to be able to create
empty test reference directories. This does not work well with many
revision control systems. As a workaround, add a file named EMPTY to
the list of ignored files in the test harness. This file can be used
as a placeholder in otherwise empty test directories.

Change-Id: I583c8c4e55479f0d48fa99d0b0d1eac9221e6652
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
8 years agobase: eliminate ipython warning
Curtis Dunham [Thu, 15 Sep 2016 17:21:38 +0000 (18:21 +0100)]
base: eliminate ipython warning

Change-Id: I3e282baeb969b6bb9534813a2f433d68246c0669
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
8 years agoarm: Add m5_fail support for aarch64
Ricardo Alves [Thu, 15 Sep 2016 17:21:24 +0000 (18:21 +0100)]
arm: Add m5_fail support for aarch64

Change-Id: Id2acbc09772be310a0eb9e33295afab07e08a4fa
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
8 years agocpu: Support exit when any one Trace CPU completes replay
Radhika Jagtap [Thu, 15 Sep 2016 17:01:20 +0000 (18:01 +0100)]
cpu: Support exit when any one Trace CPU completes replay

This change adds a Trace CPU param to exit simulation early,
i.e. when the first (any one) trace execution is complete. With
this change the user gets a choice to configure exit as either
when the last CPU finishes (default) or first CPU finishes
replay. Configuring an early exit enables simulating and
measuring stats strictly when memory-system resources are being
stressed by all Trace CPUs.

Change-Id: I3998045fdcc5cd343e1ca92d18dd7f7ecdba8f1d
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
8 years agocpu: Adjust for trace offset and fix stats
Radhika Jagtap [Thu, 15 Sep 2016 17:01:16 +0000 (18:01 +0100)]
cpu: Adjust for trace offset and fix stats

This change subtracts the time offset present in the trace from
all the event times when nodes and request are sent so that the
replay starts immediately when the simulation starts. This makes
the stats accurate when the time offset in traces is large, for
example when traces are generated in the middle of a workload
execution. It also solves the problem of unnecessary DRAM
refresh events that would keep occuring during the large time
offset before even a single request is replayed into the system.

Change-Id: Ie0898842615def867ffd5c219948386d952af7f7
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
8 years agocpu: Add frequency scaling to the Trace CPU
Radhika Jagtap [Thu, 15 Sep 2016 17:01:09 +0000 (18:01 +0100)]
cpu: Add frequency scaling to the Trace CPU

This change adds a simple feature to scale the frequency of
the Trace CPU.

The compute delays in the input traces provide timing. This
change adds a freqency multiplier parameter to the Trace CPU
set to 1.0 by default. The compute delay is manipulated to
effectively achieve the  frequency at which the nodes become
ready and thus scale the frequency of the Trace CPU.

Change-Id: Iaabbd57806941ad56094fcddbeb38fcee1172431
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
8 years agoarm, config: Fixups for the example big.LITTLE(tm) configuration
Gabor Dozsa [Thu, 15 Sep 2016 17:00:59 +0000 (18:00 +0100)]
arm, config: Fixups for the example big.LITTLE(tm) configuration

This patch refactors the configuration file to use a more
object-oriented design.

Change-Id: I44ac2d063c2b5901f385544fb6ce3f259459cb05
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Gabor Dozsa <gabor.dozsa@arm.com>
8 years agokvm: Support timing accesses for KVM cpu
Michael LeBeane [Wed, 14 Sep 2016 03:20:03 +0000 (23:20 -0400)]
kvm: Support timing accesses for KVM cpu
This patch enables timing accesses for KVM cpu.  A new state,
RunningMMIOPending, is added to indicate that there are outstanding timing
requests generated by KVM in the system.  KVM's tick() is disabled and the
simulation does not enter into KVM until all outstanding timing requests have
completed.  The main motivation for this is to allow KVM CPU to perform MMIO
in Ruby, since Ruby does not support atomic accesses.

8 years agox86: Force strict ordering for memory mapped m5ops
Michael LeBeane [Wed, 14 Sep 2016 03:18:34 +0000 (23:18 -0400)]
x86: Force strict ordering for memory mapped m5ops
Normal MMAPPED_IPR requests are allowed to execute speculatively under the
assumption that they have no side effects.  The special case of m5ops that are
treated like MMAPPED_IPR should not be allowed to execute speculatively, since
they can have side-effects.  Adding the STRICT_ORDER flag to these requests
blocks execution until the associated instruction hits the ROB head.

8 years agosim: Refactor quiesce and remove FS asserts
Michael LeBeane [Wed, 14 Sep 2016 03:17:42 +0000 (23:17 -0400)]
sim: Refactor quiesce and remove FS asserts
The quiesce family of magic ops can be simplified by the inclusion of
quiesceTick() and quiesce() functions on ThreadContext.  This patch also
gets rid of the FS guards, since suspending a CPU is also a valid
operation for SE mode.

8 years agoconfig: move dist-gem5 options to common config
Michael LeBeane [Wed, 14 Sep 2016 03:16:06 +0000 (23:16 -0400)]
config: move dist-gem5 options to common config
dist-gem5 should not be restricted to FullSystem mode.

8 years agodev: Add a DmaCallback class to DmaDevice
Michael LeBeane [Wed, 14 Sep 2016 03:14:24 +0000 (23:14 -0400)]
dev: Add a DmaCallback class to DmaDevice
This patch introduces the DmaCallback helper class, which registers a callback
to fire after a sequence of (potentially non-contiguous) DMA transfers on a
DmaPort completes.

8 years agosim, syscall_emul: Add mmap to EmulatedDriver
Michael LeBeane [Wed, 14 Sep 2016 03:12:46 +0000 (23:12 -0400)]
sim, syscall_emul: Add mmap to EmulatedDriver
Add support for calling mmap on an EmulatedDriver file descriptor.

8 years agogpu-compute: Fix bug with return in cfg
Michael LeBeane [Wed, 14 Sep 2016 03:11:20 +0000 (23:11 -0400)]
gpu-compute: Fix bug with return in cfg
Connecting basic blocks would stop too early in kernels where ret was not the
last instruction.  This patch allows basic blocks after the ret instruction
to be properly connected.

8 years agodev: Exit correctly in dist-gem5
Michael LeBeane [Wed, 14 Sep 2016 03:08:34 +0000 (23:08 -0400)]
dev: Exit correctly in dist-gem5
The receiver thread in dist_iface is allowed to directly exit the simulation.
This can cause exit to be called twice if the main thread simultaneously wants
to exit the simulation.  Therefore, have the receiver thread enqueue a request
to exit on the primary event queue for the main simulation thread to handle.

8 years agomisc: Remove FullSystem check for networking components
Michael LeBeane [Wed, 14 Sep 2016 03:06:32 +0000 (23:06 -0400)]
misc: Remove FullSystem check for networking components
Ethernet devices are currently only hooked up if running in FS mode.  Much of
the Ethernet networking code is generic and can be used to build non-Ethernet
device models.  Some of these device models do not require a complex driver
stack and can be built to use an EmulatedDriver in SE mode. This patch enables
etherent interfaces to properly connect regardless of whether the simulation
is in FS or SE mode.

8 years agobase: Output all AddrRange parameters to config.ini
Matt Poremba [Wed, 14 Sep 2016 03:06:18 +0000 (23:06 -0400)]
base: Output all AddrRange parameters to config.ini

Currently only 'start' and 'end' of AddrRange are printed in config.ini.
This causes address ranges to be overlapping when loading a c++-only
config with interleaved addresses using CxxConfigManger. This patch adds
prints for the interleave and XOR bits to config.ini such that address
ranges are properly setup with cxx config.

8 years agodev, arm: Add a customizable NoMali GPU model
Andreas Sandberg [Tue, 6 Sep 2016 09:22:38 +0000 (10:22 +0100)]
dev, arm: Add a customizable NoMali GPU model

Add a customizable NoMali GPU model and an example Mali T760
configuration. Unlike the normal NoMali model (NoMaliGpu), the
NoMaliCustopmGpu model exposes all the important GPU ID registers to
Python. This makes it possible to implement custom GPU configurations
by without changing the underlying NoMali library.

Change-Id: I4fdba05844c3589893aa1a4c11dc376ec33d4e9e
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
8 years agoext: eliminate warnings in SST connector
Curtis Dunham [Fri, 2 Sep 2016 13:58:15 +0000 (14:58 +0100)]
ext: eliminate warnings in SST connector

Now compiles completely clean.

8 years agocommit 15c633eea52f21dae8cb3a195823b3cdec7be491
Curtis Dunham [Wed, 24 Aug 2016 13:20:53 +0000 (14:20 +0100)]
commit 15c633eea52f21dae8cb3a195823b3cdec7be491
Author: Curtis Dunham <Curtis.Dunham@arm.com>
    ext: update SST connector for SST 6.0

8 years agoconfig: KVM acceleration for apu_se.py
David Hashe [Mon, 22 Aug 2016 15:43:44 +0000 (11:43 -0400)]
config: KVM acceleration for apu_se.py

Add support for using KVM to accelerate APU simulations. The intended use
case is to fast-forward through runtime initialization until the first
kernel launch.

8 years agotests: Add example of using KVM acceleration with an app
David Hashe [Mon, 22 Aug 2016 15:41:37 +0000 (11:41 -0400)]
tests: Add example of using KVM acceleration with an app

Add #ifdef's to gpu-hello.cpp demonstrating how to annotate an application
for KVM acceleration.

8 years agocpu, mem, sim: Change how KVM maps memory
David Hashe [Mon, 22 Aug 2016 15:41:05 +0000 (11:41 -0400)]
cpu, mem, sim: Change how KVM maps memory

Only map memories into the KVM guest address space that are
marked as usable by KVM. Create BackingStoreEntry class
containing flags for is_conf_reported, in_addr_map, and
kvm_map.

8 years agodev: Revert 0a316996de76 [dev, sim: Added missing override...]
Andreas Sandberg [Tue, 16 Aug 2016 09:59:15 +0000 (10:59 +0100)]
dev: Revert 0a316996de76 [dev, sim: Added missing override...]

This changeset reverts the changset "dev, sim: Added missing override
keywords to fix CLANG compilation (OSX)" which was incorrectly rebased.

Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
8 years agocpu: Add missing override in Minor's exec context
Andreas Sandberg [Mon, 15 Aug 2016 11:00:37 +0000 (12:00 +0100)]
cpu: Add missing override in Minor's exec context

Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
8 years agocpu: Fixed clang errors. Added 'override' keyword for virtual functions.
Reiley Jeapaul [Mon, 15 Aug 2016 11:00:36 +0000 (12:00 +0100)]
cpu: Fixed clang errors. Added 'override' keyword for virtual functions.

Change-Id: Ic37311443ca11ee6d95bceffea599e054e7aa110
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
8 years agomem: Print an MSHR without triggering any assertions
Nikos Nikoleris [Mon, 15 Aug 2016 11:00:36 +0000 (12:00 +0100)]
mem: Print an MSHR without triggering any assertions

Previously printing an mshr would trigger an assertion if the MSHR was
not in service or if the targets list was empty. This patch changes
the print function to bypasses the accessor functions for
postInvalidate and postDowngrade and avoid the relevant assertions. It
also checks if the targets list is empty before calling print on it.

Change-Id: Ic18bee6cb088f63976112eba40e89501237cfe62
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
8 years agodev, sim: Added missing override keywords to fix CLANG compilation (OSX)
Matteo Andreozzi [Mon, 15 Aug 2016 11:00:35 +0000 (12:00 +0100)]
dev, sim: Added missing override keywords to fix CLANG compilation (OSX)

Change-Id: Ice5fa11e77d06576eaa42149f5fa340a769d8b01
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
8 years agocpu, arch: fix the type used for the request flags
Nikos Nikoleris [Mon, 15 Aug 2016 11:00:35 +0000 (12:00 +0100)]
cpu, arch: fix the type used for the request flags

Change-Id: I183b9942929c873c3272ce6d1abd4ebc472c7132
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
8 years agotests: remove EIO tests
Steve Reinhardt [Sun, 14 Aug 2016 03:07:28 +0000 (23:07 -0400)]
tests: remove EIO tests

An email sent to gem5-users and gem5-dev asking if anyone was
still using EIO traces got no responses, so it seems like it's
not worth maintaining this any longer.

8 years agostats: Update to match classic memory changes
Andreas Sandberg [Fri, 12 Aug 2016 13:12:59 +0000 (14:12 +0100)]
stats: Update to match classic memory changes

8 years agomem: Add support for secure packets in the snoop filter
Nikos Nikoleris [Fri, 12 Aug 2016 13:11:45 +0000 (14:11 +0100)]
mem: Add support for secure packets in the snoop filter

Secure and non-secure data can coexist in the cache and therefore the
snoop filter should treat differently packets with secure and non
secure accesses. This patch uses the lower bits of the line address to
keep track of whether the packet is addressing secure memory or not.

Change-Id: I54a5e614dad566a5083582bede86c86896f2c2c1
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Stephan Diestelhorst <stephan.diestelhorst@arm.com>
Reviewed-by: Tony Gutierrez <anthony.gutierrez@amd.com>
8 years agomem: Add snoop filter to SystemXBar by default
Andreas Hansson [Fri, 12 Aug 2016 13:11:45 +0000 (14:11 +0100)]
mem: Add snoop filter to SystemXBar by default

This patch changes the default behaviour of the SystemXBar, adding a
snoop filter. With the recent updates to the snoop filter allocation
behaviour this change no longer causes problems for the regressions
without caches.

Change-Id: Ibe0cd437b71b2ede9002384126553679acc69cc1
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Tony Gutierrez <anthony.gutierrez@amd.com>
8 years agomem: Use FromCache attribute in snoop filter allocation
Andreas Hansson [Fri, 12 Aug 2016 13:11:45 +0000 (14:11 +0100)]
mem: Use FromCache attribute in snoop filter allocation

This patch improves the snoop filter allocation decisions by not only
looking at whether a port is snooping or not, but also if the packet
actually came from a cache. The issue with only looking at isSnooping
is that the CPU ports, for example, are snooping, but not actually
caching. Previously we ended up incorrectly allocating entries in
systems without caches (such as the atomic and timing quick
regressions). Eventually these misguided allocations caused the snoop
filter to panic due to an excessive size.

On the request path we now include the fromCache check on the packet
itself, and for responses we check if we actually have a snoop-filter
entry.

Change-Id: Idd2dbc4f00c7e07d331e9a02658aee30d0350d7e
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Stephan Diestelhorst <stephan.diestelhorst@arm.com>
Reviewed-by: Tony Gutierrez <anthony.gutierrez@amd.com>
8 years agomem: Update mostly exclusive policy even further
Andreas Hansson [Fri, 12 Aug 2016 13:11:45 +0000 (14:11 +0100)]
mem: Update mostly exclusive policy even further

This patch takes yet another step in maintaining the clusivity, in
that it allows a mostly-inclusive cache to hold on to blocks even when
responding to a ReadExReq or UpgradeReq. Previously the cache simply
invalidated these blocks, but there is no strict need to do so.

The most important part of this patch is that we simply mark the block
clean when satisfying the upstream request where the cache is allowed
to keep the block. The only tricky part of the patch is in the memory
management of deferred snoops, where we need to distinguish the cases
where only the packet was copied (we expected to respond), and the
cases where we created an entirely new packet and request (we kept it
only to replay later).

The code in satisfyRequest is definitely ready for some refactoring
after this.

Change-Id: I201ddc7b2582eaa46fb8cff0c7ad09e02d64b0fc
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Tony Gutierrez <anthony.gutierrez@amd.com>
8 years agomem: Update mostly exclusive cache policy to cover more cases
Andreas Hansson [Fri, 12 Aug 2016 13:11:45 +0000 (14:11 +0100)]
mem: Update mostly exclusive cache policy to cover more cases

This patch changes how the mostly exclusive policy is enforced to
ensure that we drop blocks when we should. As part of this change, the
actual invalidation due to the clusivity enforcement is moved outside
the hit handling, to a separate method maintainClusivity. For the
timing mode that means we can deal with all MSHR targets before taking
any action and possibly dropping the block. The method
satisfyCpuSideRequest is also renamed satisfyRequest as part of this
change (since we only ever see requests from the cpu-side port).

Change-Id: If6f3d1e0c3e7be9a67b72a55e4fc2ec4a90fd3d2
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Tony Gutierrez <anthony.gutierrez@amd.com>
8 years agomem: Add a FromCache packet attribute
Andreas Hansson [Fri, 12 Aug 2016 13:11:45 +0000 (14:11 +0100)]
mem: Add a FromCache packet attribute

This patch adds a FromCache attribute to the packet, and updates a
number of the existing request commands to reflect that the request
originates from a cache. The attribute simplifies checking if a
requests came from a cache or not, and this is used by both the cache
and snoop filter in follow-on patches.

Change-Id: Ib0a7a080bbe4d6036ddd84b46fd45bc7eb41cd8f
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Tony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Steve Reinhardt <stever@gmail.com>
8 years agoarm, config: Exit with fatal error if using Ruby
Andreas Sandberg [Wed, 10 Aug 2016 15:40:14 +0000 (16:40 +0100)]
arm, config: Exit with fatal error if using Ruby

Ruby on ARM is currently very experimental. Fail with a fatal error
that explains this to make sure users are aware of the limitations (it
doesn't actually work yet!).

Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
8 years agoarm, config: Add initial support for Ruby
Andreas Sandberg [Wed, 10 Aug 2016 15:26:34 +0000 (16:26 +0100)]
arm, config: Add initial support for Ruby

Add initial support for creating an ARM system with a Ruby-based
memory system. This support is currently experimental and limited to
the new VExpress_GEM5_V1 platform.

Change-Id: I36baeb68b0d891e34ea46aafe17b5e55217b4bfa
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Brad Beckmann <brad.beckmann@amd.com>
8 years agoarm, dev: Add support for listing DMA ports in new platforms
Andreas Sandberg [Wed, 10 Aug 2016 14:27:26 +0000 (15:27 +0100)]
arm, dev: Add support for listing DMA ports in new platforms

When using a Ruby memory system, the Ruby configuration scripts expect
to get a list of DMA ports to create the necessary DMA sequencers. Add
support in the utility functions that wire up devices to append DMA
ports to a list instead of connecting them to the IO bus. These
functions are currently only used by the VExpress_GEM5_V1 platform.

Change-Id: I46059e46b0f69e7be5f267e396811bd3caa3ed63
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-by: Brad Beckmann <brad.beckmann@amd.com>
8 years agoruby: Implement support for functional accesses to PIO ranges
Andreas Sandberg [Wed, 10 Aug 2016 14:27:13 +0000 (15:27 +0100)]
ruby: Implement support for functional accesses to PIO ranges

There are cases where we want to put boot ROMs on the PIO bus. Ruby
currently doesn't support functional accesses to such memories since
functional accesses are always assumed to go to physical memory. Add
the required support for routing functional accesses to the PIO bus.

Change-Id: Ia5b0fcbe87b9642bfd6ff98a55f71909d1a804e3
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Brad Beckmann <brad.beckmann@amd.com>
Reviewed-by: Michael LeBeane <michael.lebeane@amd.com>
8 years agoarm: Don't report the boot ROM as a memory in config tables
Andreas Sandberg [Wed, 10 Aug 2016 13:49:11 +0000 (14:49 +0100)]
arm: Don't report the boot ROM as a memory in config tables

The boot ROM shouldn't be used as a memory by the kernel. Memories
have a flag to indicate this which is set for some platforms. Update
all platforms to consistently set this flag to indicate that the boot
ROM shouldn't be reported as normal memory.

Change-Id: I2bf0273e99d2a668e4e8d59f535c1910c745aa7b
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Brad Beckmann <brad.beckmann@amd.com>
--HG--
extra : amend_source : c2cbda38636ea37cbe9ae6977a06b923eab5ba56

8 years agosim: fix issues with pwrite(); don't enable fstatfs
Tony Gutierrez [Fri, 5 Aug 2016 21:15:19 +0000 (17:15 -0400)]
sim: fix issues with pwrite(); don't enable fstatfs

this patch fixes issues with changeset 11593

use the host's pwrite() syscall for pwrite64Func(),
as opposed to pwrite64(), because pwrite64() does
not work well on all distros.

undo the enabling of fstatfs, as we will add this
in a separate pate.

8 years agox86, sim: add some syscalls to X86
Tony Gutierrez [Thu, 4 Aug 2016 16:32:21 +0000 (12:32 -0400)]
x86, sim: add some syscalls to X86

this patch adds an implementation for the pwrite64 syscall and
enables it for x86_64, and enables fstatfs for x86_64.

8 years agostyle: Make the style fixers safe
Jason Lowe-Power [Wed, 3 Aug 2016 16:10:46 +0000 (11:10 -0500)]
style: Make the style fixers safe

Adds a wrapper to the fix functions of the verifiers. This wrapper first
copies the original file to a backup file, then performs the fix. If an
error occurs, the backup file is used to restore the original file.

Also fixed a line-length error in verifiers.py