NOTE: With this change there is a possibility for `DRAMCtrl::Rank`s
event names to not properly match the rank they were generated by. This
could occur if the public rank member is modified after the Rank's
construction. A patch would mean refactoring Rank and `DRAMCtrl`b to
privatize many of the members of Rank behind getters.
Change-Id: I7b8bd15086f4ffdfd3f40be4aeddac5e786fd78e
Signed-off-by: Sean Wilson <spwilson2@wisc.edu>
Reviewed-on: https://gem5-review.googlesource.com/3745
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
In order to replicate the same `name()` output with `PacketQueue`, subclasses
using EventFunctionWrapper must initialize PacketQueue with their own name so
the sendEvent holds the name of the subclass.
Change-Id: Ib091e118bab8858192e1d1370d61def42958ec29
Signed-off-by: Sean Wilson <spwilson2@wisc.edu>
Reviewed-on: https://gem5-review.googlesource.com/3744
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
This change was made so Rank objects have their name assigned
when they are instantiated. Therefore, they can initialize their
member objects with their name and it is less likely to change during
runtime.
(NOTE: I would recommend hiding the fields which would cause the name to
change behind getters. Since modification of `Rank.rank` during runtime
will cause the `name()` to change.)
Change-Id: Id51c3553b40e489792c57950e18b8ce927e43173
Signed-off-by: Sean Wilson <spwilson2@wisc.edu>
Reviewed-on: https://gem5-review.googlesource.com/3742
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
This commit adds support for diod to use a unix socket, rather
than a TCP port. We don't rely on the IP-based connection as we
directly use pipes to interact with diod. This allows it to work
on any system, even if the specific port is taken by another diod
instance (or similar). Secondly, the Unix socket could in theory
be used to debug. However, this functionality has not been
tested.
Change-Id: I616e0ad8768da1dfc867de3af98cdfbb22a72d63
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Sascha Bischoff <sascha.bischoff@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2820
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
GDB breaks if more bytes are sent than the transmitted registers
actually need. Therefore the GdbRegCache struct needs to be packed to
prevent padding at the end.
Change-Id: Ib2c14eb70becdac609eb4f475d5dddbd5bcc60da
Signed-off-by: Matthias Hille <matthiashille8@gmail.com>
Reviewed-on: https://gem5-review.googlesource.com/3020
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
This patch fix some statistics that in presence of a resetStats
instruction were not reseted. This bug makes impossible to obtain
reliable network statistics when the simulation doesn't start from tick
zero.
Change-Id: Ibec45f08d95bf0a533d94b70ec960719206ae945
Maintainer: Tushar Krishna <tushar@ece.gatech.edu>
Reviewed-on: https://gem5-review.googlesource.com/3700
Reviewed-by: Jieming Yin <bjm419@gmail.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Previously the directory covered a flat address range that always
started from address 0. This change adds a vector of address ranges
with interleaving and hashing that each directory keeps track of and
the necessary flexibility to support systems with non continuous
memory ranges.
Change-Id: I6ea1c629bdf4c5137b7d9c89dbaf6c826adfd977
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2903
Reviewed-by: Bradford Beckmann <brad.beckmann@amd.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
When it comes time to send a packet with the i8254xGBe model hooked up to
EtherTap and while running in KVM mode, the packet will first go to the
EtherTap over the network style port between them. EtherTap, because it's
not actually a model of anything in the simulation, will immediately pass
the packet off to the real network and report that the transmission was
successful to the i8254xGBe. The i8254xGBe will notice that it still has
stuff it can send (the KVM mode CPU has no trouble keeping it full) and
will, without returning and collapsing the stack, immediately call back
into EtherTap with the next packet. This loop repeats, continually
deepening the stack, until gem5 crashes with a segfault.
To break this loop, a few small changes have been made. First, txFifoTick
has been repurposed slightly so that it continuously keeps track of
whether there's still work to do to flush out the fifo during the current
tick. The code in txWire has been adjusted slightly so that it clears that
variable at the start (also removing some redundancy), so that other code
can set it again if more work needs to be done. Specifically, the
ethTxDone function will set that flag.
If there's more work to be done flushing the Fifo while in tick(), it
will loop until txFifoTick stays cleared, meaning either the Fifo is
empty, or the object on the other end hasn't said it's done yet.
Finally, a new bool member called inTick has been added which keeps track
of whether the tick() function is still active somewhere in the callstack.
If it is, then the tick event shouldn't be rescheduled in ethTxDone, since
tick will take care of that before it returns. It won't check to see if it
needs to, and so without this check there's a panic from scheduling the
same event twice.
It's not completely clear that the Fifo should send packets over and over
as fast as the other side accepts them within the same tick, although it's
not clear that it shouldn't either. If not, then probably all that would
need to change would be to remove the "while" loop so that the tick event
would be rescheduled, and the Fifo would be further emptied the next time
around.
Change-Id: I653379b43389d0539ecfadb3fc6c40e38a8864c2
Reviewed-on: https://gem5-review.googlesource.com/3642
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
Maintainer: Gabe Black <gabeblack@google.com>
gcc has had a lot of problems with incremental linking and partial linking
at the same time. Basically, the partial link assumes that it's the only
link that's going to happen, and it converts weak external symbols into
regular external symbols. Then when the real final link happens, those
symbols are duplicated and the link fails.
Versions of gcc 6 and greater add an option called -flinker-output which
lets you tell the linker to do an incremental link. Unfortunately, other
bugs make that fail, and so gcc 6 doesn't work either. Hopefully version
7 works better.
A --force-lto option was added so that, when only one of lto and partial
linking is available, you can switch from having partial linking to having
lto.
Change-Id: I5e293f5cfb07a14343dc74030d99cb161fb8bbbe
Reviewed-on: https://gem5-review.googlesource.com/3680
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Some of the macros, notably panic, uses exit(). Callers shouldn't have to
know that or have coincidentally included cstdlib, the provider of exit,
themselves.
Change-Id: I634602ed1795dcc8897b4bddb1167c96763acc18
Reviewed-on: https://gem5-review.googlesource.com/3601
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
The object is called EtherTap (as opposed to EtherTapStub, what the former
EtherTap was renamed to), and its existance is gated on the linux/if_tun.h
header file existing. That's probably overly strict, but it will hopefully
be minimally likely to break the build for other systems.
Change-Id: Ie03507fadf0d843a4d4d52f283c44a416c6f2a74
Reviewed-on: https://gem5-review.googlesource.com/3646
Reviewed-by: Nathan Binkert <nate@binkert.org>
Maintainer: Nathan Binkert <nate@binkert.org>
A lot of the implementation of EtherTapStub can be shared with a version
which uses a tap device directly. This change factors out those parts to
accommodate that.
Change-Id: I9c2e31f1be139ca73859a83f05457cef90101006
Reviewed-on: https://gem5-review.googlesource.com/3645
Reviewed-by: Nathan Binkert <nate@binkert.org>
Maintainer: Nathan Binkert <nate@binkert.org>
The EtherTap object is going to be reworked so that it connects to a tap
device directly, but it's worthwhile to still be able to use the m5tap
utility (util/tap) to send/receive packets on systems which don't support
tap but do support the pcap API. It can also be used to replay ethernet
frames, to capture the ethernet frames coming from gem5 for analysis, to
programmatically consume and/or generate the frames, or even to forward
them to/from a remote system.
Change-Id: Ic7bd763d86cd913ac373dd10a8d6d1fc6b35f95a
Reviewed-on: https://gem5-review.googlesource.com/3644
Reviewed-by: Nathan Binkert <nate@binkert.org>
Maintainer: Nathan Binkert <nate@binkert.org>
There needs to be a SlavePort called "tap" for the ethertap device to be
able to connect to the gem5 network successfully.
Change-Id: I1ad81219f612fd1ec278c6148af728d20bc916da
Reviewed-on: https://gem5-review.googlesource.com/3580
Reviewed-by: Nathan Binkert <nate@binkert.org>
Maintainer: Nathan Binkert <nate@binkert.org>
Previously the Sequencer upon a Store Conditional would
unconditionally set the data of the memory location. This change
checks and prevents a failed Store Conditional from modifying any
data.
Change-Id: Id63c9579d8f054f0e95c6d338a7e31aa48762755
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2902
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Bradford Beckmann <brad.beckmann@amd.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Previously proxy vector parameters would resolve correctly only for
Parent.all. Any other proxy such as Parent.any, or exact ones such as
Parent.addr_range would resolve to a *vector* of the right value
resulting into a vector of a vector. For example if we set:
DirectoryController0.addr_range = [0x100000-0x1fffff, 0x200000-0x2fffff]
DirectoryMemory0.addr_range = Parent.addr_range
where DirectoryController0 is the parent SimObject of DirectoryMemory0
after unproxying the Parent.addr_range VectorParam we would get
DirectoryMemory0.addr_range = [[0x100000-0x1fffff, 0x200000-0x2fffff]]
This change unifies handling of all three proxies to the same correct
unproxy mechanism.
Change-Id: Ie5107f69f58eb700b3e1b92c55210e0d53e6788d
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2901
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Remove redundant information from the ExtMachInst, hash the vex
information to ensure the decode cache works properly, print the vex info
when printing an ExtMachInst, consider the vex info when comparing two
ExtMachInsts, fold the info from the vex prefixes into existing settings,
remove redundant decode code, handle vex prefixes one byte at a time and
don't bother building up the entire prefix, and let instructions that care
about vex use it in their implementation, instead of developing an entire
parallel decode tree.
This also eliminates the error prone vex immediate decode table which was
incomplete and would result in an out of bounds access for incorrectly
encoded instructions or when the CPU was mispeculating, as it was (as far
as I can tell) redundant with the tables that already existed for two and
three byte opcodes. There were differences, but I think those may have
been mistakes based on the documentation I found.
Also, in 32 bit mode, the VEX prefixes might actually be LDS or LES
instructions which are still legal in that mode. A valid VEX prefix would
look like an LDS/LES with an otherwise invalid modrm encoding, so use that
as a signal to abort processing the VEX and turn the instruction into an
LES/LDS as appropriate.
Change-Id: Icb367eaaa35590692df1c98862f315da4c139f5c
Reviewed-on: https://gem5-review.googlesource.com/3501
Reviewed-by: Joe Gross <joe.gross@amd.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
When the LiveProcess class was renamed to be just Process, the CL author
also changed the syscall function from a virtual function into a regular
one. Unfortunately, the I386Process class overrode the syscall function
to adjust the return address so that control would return to the right
place. Without that adjustment, 32 bit x86 process would segfault and die
immediately after their first system call.
This change reinstates the virtual specifier on the base syscall function,
and adds an override keyword on the I386Process's version so that it won't
be orphaned again in the future. It also fixes some small style issues the
style checker script complained about.
Change-Id: I0d1178ea0eda6676050c8fc043820a2bb4d99c0d
Reviewed-on: https://gem5-review.googlesource.com/3500
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Brandon Potter <Brandon.Potter@amd.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
The current implementation of reference counting for PyEvents only
partially works. The native object is currently kept alive while it is
in the event queue. However, if the Python object goes out of scope,
the Python side of this object is garbage collected which leaves a
"dangling" native object. This results in confusing error messages
where PyBind is unable to find the Python implementation of an event
when it is triggered.
Implement reference counting using the generalized reference counting
API instead.
Change-Id: I4e8e04abc4f61dff238d718065f5371e73b38ab3
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/3222
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
We currently only support deleting an event if it is triggered and not
re-scheduled. This is fine for most native code. However, there are
cases where Python needs to count references to make sure that the
Python object stays live while the native object is live.
Generalise the mechanism used to implement by adding reference
counting hooks to the event base class:
* Event::acquire() / Event::acquireImpl()
* Event::release() / Event::releaseImpl()
These calls can be used to implement both reference counting and the
existing AutoDelete functionality. The default implementation in Event
maintains backwards compatibility with the existing AutoDelete feature
by ignoring acquireImpl() and deleting the event on releaseImpl() if
it isn't scheduled anymore.
Since AutoDelete functionality is no longer the only way events can be
managed, this change introduces the new Managed flag. This flag
activates automatic memory management. The acquireImpl()/releaseImpl()
methods are only called from acquire()/release() it is set. To
maintain backwards compatibility, AutoDelete is used as an alias for
Managed.
Change-Id: I5637984c906a9d44c22780712cf1c521b8297149
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/3221
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
This patch fixes a problem with RISC-V initial stack setup in SE mode
where the AT_RANDOM aux vector value contains an address that is too
close to the top of the stack and doesn't fit the required 16 bytes. To
fix this, the program header table was added to the top of the stack
just like the RISC-V proxy kernel does.
Change-Id: I814562e060ff041cd0d7a7c54c3685645bd325a3
Reviewed-on: https://gem5-review.googlesource.com/3401
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Brandon Potter <Brandon.Potter@amd.com>
Maintainer: Alec Roelke <ar4jc@virginia.edu>
The ISA code for ARM calculates min and max elements for types using
bit manipulation. That triggers some warnings, treated as errors, as
the compiler can tell that there is an overflow and the sign
flips. Fixed using standard lib definitions instead.
Change-Id: Ie2331b410c7f76d4bd87da5afe9edf20c8ac91b3
Reviewed-by: Giacomo Gabrielli <giacomo.gabrielli@arm.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/3481
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
The PyBind wrappers could potentially delete SimObjects if they don't
have any references. This is not desirable since there could be
pointers to such objects within the C++ world. This problem doesn't
normally occur since Python typically holds a pointer to the root node
as long as the simulator is running.
Prevent SimObject and Param deletion by using a PyBind-prescribed
unique_ptr with a dummy deleter as the pointer wrapper for the Python
world.
Change-Id: Ied14602c9ee69a083a69c5dae1b5fcf8efb4548a
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/3224
Reviewed-by: Gabe Black <gabeblack@google.com>
There is a weird issue with the PyBind wrapper of
vector<AddrRange>. Assigning new values to a param that is a vector of
AddrRange sometimes results in an out-of-bounds memory access.
We work around this issue by treating AddrRange vectors as opaque
types. This slightly changes the semantics of the wrapper since Python
now manipulates the real object rather than a copy that has been
converted to a list.
Change-Id: Ie027c06e7a7262214b43b19a76b24fe4b20426c5
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Sascha Bischoff <sascha.bischoff@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-by: Timothy Hayes <timothy.hayes@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/3223
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Compiling gem5 with recent version of clang (4 and 5) triggers
warnings that are treated as errors:
* Global templatized static functions result in a warning if they
are not used. These should either be declared as static inline or
without the static identifier to avoid the warning.
* Some templatized classes contain static variables. The
instantiated versions of these variables / templates need to be
explicitly declared to avoid a compiler warning.
Change-Id: Ie8261144836e94ebab7ea04ccccb90927672c257
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/3420
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
The new version modularizes the implementation of the various commands,
gets rid of dynamic allocation of the register cache, fixes some small
style problems, and uses exceptions to simplify error handling internal to
the GDB stub.
Change-Id: Iff3548373ce4adfb99106a810f5713b769df89b2
Reviewed-on: https://gem5-review.googlesource.com/3280
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Boris Shingarov <shingarov@gmail.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
If the operands were 64 bit, an intermediate calculation could lose a
carry bit. This change rearranges that intermediate calculation if the
operand width is large, and reworks the microop implementation in general
in an attempt to make it easier to understand.
Change-Id: Ib36333f3f2695a33cd9623e43682de22ebd2e7ea
Reviewed-on: https://gem5-review.googlesource.com/3381
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Add a helper function, m5.event.create(), to create events from
Python. This function takes a callable Python object (e.g., a
function) as an argument and optionally a priority as a keyword
argument. This function was accidentally dropped from the public API
when switching to PyBind.
Change-Id: Icbd0e392d9506934ec2c9f541199aa35c1c2df8c
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/3220
Reviewed-by: Gabe Black <gabeblack@google.com>
The style checker complains about line length and ordering for these
files. This fix should make these two files kosher.
Change-Id: I822a0518a98d9e379a543d2017e90c4e9666a58d
Reviewed-on: https://gem5-review.googlesource.com/3380
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Brandon Potter <Brandon.Potter@amd.com>
If a (regular) store is followed closely enough by a locked load that
overlaps, the LSQ will forward the store's data to the locked load and
never tell the cache about the locked load. As a result, the cache will
not lock the address and all future store-conditional requests on that
address will fail. This patch fixes that by preventing forwarding if
the memory request is a locked load and adding another case to the LSQ
forwarding logic that delays the locked load request if a store in the
LSQ contains all or part of the data that is requested.
[Merge second and last if blocks because their bodies are the same.]
Change-Id: I895cc2b9570035267bdf6ae3fdc8a09049969841
Reviewed-on: https://gem5-review.googlesource.com/2400
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Tony Gutierrez <anthony.gutierrez@amd.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
We don't implement the GICD_IGROUPRn registers, which is allowed, but
to be correct, they should be RAZ/WI (read as zero, writes ignored).
Change-Id: I8039baf72f45c0095f41e165b8e327c79b1ac082
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2620
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Currently, if the remote gdb stub fails to read a byte from an incoming
packet because the connection has been dropped, the read call will return
anyway and the calling code will have no way to know something bad
happened. It might reattempt the read over and over again waiting for some
particular byte, doomed to never make forward progress.
This change modifies the remote GDB code so that if a read or write call
fails, it will instead detach from the debugger and continue. Before this
change, When simulating a port scan, ie connecting to the debugger port
and then immediately dropping the connection using this command:
nc -v -n -z -w 1 127.0.0.1 7000
gem5 would enter the previously described death spiral. After it, gem5
detaches from the bad connection and resumes execution. Subsequently
attaching with gdb was successful.
This code is written in a C centric style, and would benefit from some
refactoring.
Change-Id: Ie3c0bb35b9cfe3671d0f731e3907548bae0d292f
Reviewed-on: https://gem5-review.googlesource.com/3180
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Clang's UBSAN implementation complains about macros that expand to
expressions that contain 'defined'. Explicitly set the various feature
macros to 1 or 0 to avoid this issue.
Change-Id: Iba239dacfe526c43ab9c5da5183a350fc4fdc57d
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-on: https://gem5-review.googlesource.com/3241
