| Age | Commit message (Collapse) | Author |
|---|
|
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>
|
|
Change-Id: I0ed4e528cb750a323facdc811dde7f0ed1ff228e
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
|
|
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>
|
|
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>
|
|
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>
|
|
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.
|
|
This changeset add fields to the process object and adds the following
three system calls: setpgid, gettid, getpid.
|
|
|
|
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.
|
|
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.
|
|
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>
|
|
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>
|
|
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>
|
|
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.
|
|
This changeset add spaces in a few spots and removes an unnecessary comment.
|
|
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.
|
|
|
|
|
|
Change-Id: I6149290d6d2ac1a4bd6165871c93d7b7d6a980ad
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
|
|
Change-Id: I29f45733c5fad822bdd0d8dcc7939d86b2e8c97b
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
|
|
Change-Id: I79c2662fc81630ab321db8a75be6cd15fa07d372
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
|
|
Change-Id: If9ebb8488e8db587482ecfa99d2c12cfe5734fb9
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
|
|
Change-Id: I4f3c2c027b1acaaf791a4c71086f34a9b9fbf4df
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
|
|
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>
|
|
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>
|
|
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>
|
|
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>
|
|
Change-Id: I0e373536897aa5bb4501b00945c2a0836100ddf4
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
|
|
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>
|
|
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.
|
|
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.
|
|
Fix compilation errors due to missing include.
|
|
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.
|
|
|
|
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.
|
|
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.
|
|
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.
|
|
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.
|
|
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.
|
|
|
|
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/
|
|
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/
|
|
Assertion in the respondEvent erroneously fired.
The assertion verifies that the controller has not moved to a low-power
state prior to receiving read data from the memory.
The original assertion triggered if the state was not:
PWR_IDLE or PWR_ACT.
In the case that failed, a periodic refresh event occurred around the
read. The REF is stalled until the final read burst is issued
and the subsequent PRE closes the bank. While the PRE will temporarily
move the state to PWR_IDLE, state will immediately transition to PWR_REF
due to the pending refresh operation. This state does not match the
assertion, which is subsequently triggered.
Fixed the assertion by explicitly checking that the state is not a low
power state
!PWR_SREF && !PWR_PRE_PDN && !PWR_ACT_PDN
Change-Id: I82921a733bbeac2bcb5a487c2f981448d41ed50b
Reviewed-by: Radhika Jagtap <radhika.jagtap@arm.com>
|
|
KvmGic functionality has been subsumed within the new MuxingKvmGic
model, which has Pl390 fallback when not using KVM for fast emulation.
This simplifies configuration and will enable checkpointing between
KVM emulation and full-system simulation.
Change-Id: Ie61251720064c512843015c075e4ac419a4081e8
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
|
|
Automatically use the MuxingKvmGic in the VExpress_GEM5_V1
platform. This removes the need to patch the host kernel or the
platform configuration when using KVM on ARM.
Change-Id: Ib1ed9b3b849b80c449ef1b62b83748f3f54ada26
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
|
|
This device allows us to, when KVM support is detected and compiled in,
instantiate the same Gic device whether the actual simulation is with
KVM cores or simulated cores. Checkpointing is not yet supported.
Change-Id: I67e4e0b6fb7ab5058e52c933f4f3d8e7ab24981e
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
|
|
A KVM VM is typically a child of the System object already, but for
solving future issues with configuration graph resolution, the most
logical way to keep track of this object is for it to be an actual
parameter of the System object.
Change-Id: I965ded22203ff8667db9ca02de0042ff1c772220
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
|
|
Change-Id: Ifc65d42eebfd109c1c622c82c3c3b3e523819e85
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
|
|
Names of DRAM configurations were updated to reflect both
the channel and device data width.
Previous naming format was:
<DEVICE_TYPE>_<DATA_RATE>_<CHANNEL_WIDTH>
The following nomenclature is now used:
<DEVICE_TYPE>_<DATA_RATE>_<n>x<w>
where n = The number of devices per rank on the channel
x = Device width
Total channel width can be calculated by n*w
Example:
A 64-bit DDR4, 2400 channel consisting of 4-bit devices:
n = 16
w = 4
The resulting configuration name is:
DDR4_2400_16x4
Updated scripts to match new naming convention.
Added unique configurations for DDR4 for:
1) 16x4
2) 8x8
3) 4x16
Change-Id: Ibd7f763b7248835c624309143cb9fc29d56a69d1
Reviewed-by: Radhika Jagtap <radhika.jagtap@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
|
|
The rr arbiter pointer in garnet was getting updated on every request,
even if there is no grant. This was leading to a huge variance in wait
time at a router at high injection rates.
This patch corrects it to update upon a grant.
|
