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>
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>
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>
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>
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
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.
Factored out of the larger banked register change.
Change-Id: I947dbdb9c00b4678bea9d4f77b913b7014208690
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Updated according to GICv2 documentation.
Change-Id: I5d926d1abf665eecc43ff0f7d6e561e1ee1c390a
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Introduce and use a lookup table.
Using fetchDescriptor() rather than DMA cleanly handles nested paging.
Change-Id: I69ec762f176bd752ba1040890e731826b58d15a6
During host bootup, KVM reads/writes to CNTHCTL_EL2. Because this
miscreg has not been implemented, the simulation would end there. This
patch causes the simulation to warn about the read/write instead of fail.
Change-Id: If034bfd0818a9a5e50c5fe86609e945258c96fa3
This fixes a bug where stage 2 lookups used the AArch32
permissions rules even if we were executing in AArch64 mode.
Change-Id: Ia40758f0599667ca7ca15268bd3bf051342c24c1
This patch restricts trapping to hypervisor only if we are in the
correct exception level for the trap to happen.
Change-Id: I0a382b6a572ef835ea36d2702b8a81b633bd3df0
Faults that could potentially be routed to the hypervisor checked
whether or not they were in a secure state without checking if security
was enabled or not. This caused faults not to be routed correctly. This
patch causes secure state checking to first ask if security is enabled.
Change-Id: I179e9b181b27f552734c9bab2b18d05ac579a119
We recompute if we are doing a stage 2 walk inside of the table walker
but we have already figured it out in the tlb. Pass the information in
to the walk instead of recomputing it.
Change-Id: I39637ce99309b2ddbc30344d45ac9ebf6a203401
The functional case is already handled within the fetchDescriptor()
function. We can thus use that function for both atomic and functional
mode when we start the table walk.
Change-Id: Iacaed28cd9024d259fd37a58150efd00ff94d86e
This patch adds the option for faults to be routed to the hypervisor
using the pre-existing routeToHyp() functions that are present in each
fault type.
Change-Id: I9735512c094457636b9870456a5be5432288e004
During address translation instructions (such as AT S1E1R_Xt) the exception
level can be different than the current exception level. This patch fixes
how the TLB determines what EL to use during these instructions.
Change-Id: Ia9ce229404de9e284bc1f7479fd2c580efd55f8f
This patch adds the AArch64 instruction hvc which raises an exception
from EL1 into EL2. The host OS uses this instruction to world switch
into the guest.
Change-Id: I930ee43f4f0abd4b35a68eb2a72e44e3ea6570be
The behavior of WFI is to cause minor to cease evaluating
pipeline logic until an interrupt is observed, however
a user may wish to drain the system while a core is sleeping
due to a WFI. This patch makes WFI drain. If an actual
drain occurs during a WFI, the CPU is already drained and will
immediately be ready for swapping, checkpointing, etc. This
should not negatively impact performance as WFI instructions
are 'stream-changing' (treated like unpredicted branches), so
all remaining instructions are wrong-path and will be squashed
rapidly.
Change-Id: I63833d5acb53d8dde78f9f0c9611de0ece385e45
This patch adds SMT support to the MinorCPU. Currently
RoundRobin or Random thread scheduling are supported.
Change-Id: I91faf39ff881af5918cca05051829fc6261f20e3
Make it so that getInterrupt *always* returns an interrupt if
checkInterrupts() returns true. This fixes/simplifies handling
of interrupts on the SMT FS CPUs (currently minor).
This patch adds a total() function to the Vector2D
stat type. Similar to other stats such as Scalar or
Vector it is useful to be able to read the total for
a given stat.
This patch fixes a bug in etherswitch. When a packet gets inserted
in the output fifo, the txEvent has to always be reschedule,
not only when an event is already scheduled. This can raise
the assertion in the reschedule function.
Don't consult the TLB test interface for PA's returned by functional
translations by the AT instruction. We implement this by chaning the
ISA code to synthesize 0-length functional reads for the TLB lookup.
The TLB then bypasses the final PA check in the tester if the size is
zero.
Change-Id: I2487b7f829cea88c37e229e9fc7a4543aced961b
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
The ELF loader currently has an assertion that checks if the size of a
loaded .text secion is non-zero. This is useful in the general case as
an empty text section normally indicates that there is something
strange with the ELF file. However, asserting isn't very useful. This
changeset converts the assert into a warning that tells the user that
something strange is happening.
Change-Id: I313e17847b50a0eca00f6bd00a54c610d626c0f0
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Currently garnet will not run due to double statistic registration of new
stats in ClockedObject. This occurs because a temporary array named 'cls'
is being added as a child to garnet internal and external link SimObjects.
This patch simply renames the temporary array which prevents it from
being added as a child object and avoids the assertion that a statistic
was already registered.
Committed by Jason Lowe-Power <jason@lowepower.com>
Sync DRAMPower to external tool
This patch syncs the DRAMPower library of gem5 to the external
one on github (https://github.com/ravenrd/DRAMPower) of which
I am a maintainer.
The version used is the commit:
902a00a1797c48a9df97ec88868f20e847680ae6
from 07. May. 2016.
Committed by Jason Lowe-Power <jason@lowepower.com>
In this new hmc configuration we have used the existing components in gem5
mainly [SerialLink] [NoncoherentXbar]& [DRAMCtrl] to define 3 different
architecture for HMC.
Highlights
1- It explores 3 different HMC architectures
2- It creates 4-HMC crossbars and attaches 16 vault controllers with it.
This will connect vaults to serial links
3- From the previous version, HMCController with round robin funtionality
is being removed and all the serial links are being accessible directly
from user ports
4- Latency incorporated by HMCController (in previous version) is being
added to SerialLink
Committed by Jason Lowe-Power <jason@lowepower.com>
This patch changes how the embedded swig code is loaded to ensure that
gem5 works with swig 3.0.9. For Python 2.7 and above, swig 3.0.9 now
relies on importlib, and actually looks in the appropriate packages,
even for the wrapped C code. However, the swig wrapper does not
explicitly place the module in the right package (it just calls
Py_InitModule), and we have to take explicit action to ensure that the
swig code can be loaded. This patch adds the information to the
generated wrappers and the appropriate calls to set the context as
part of the swig initialisation.
Previous versions of swig used to fall back on looking in the global
namespace for the wrappers (and still do for Python 2.6), but
technically things should not work without the functionality in this
patch.
Previously when we initialized the TLB we would allocate a number of
TLB entries which would be marked as valid. As a result the TLB
contained an entry which would be considered a valid entry for the 0
page.
Change-Id: I23ace86426a171a4f6200ebeb29ad57c21647036
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
The use of array tuples, requires an explicit include of the array library
Change-Id: I06730051777a97edf80e41a5604184b387b12239
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
The snoop filter handles requests in two steps which preceed and
follow the call to send the packet downstream. An address mapper could
possibly change the address of the packet when it is sent downstream
breaking the snoop filter assumption that the address is unchanged
Change-Id: Ib2db755e9ebef4f2f7c0169a46b1b11185ffbe79
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
The traffic generator currently resolves relative trace paths relative
to gem5's current working directory. This can lead to surprising
results for relative paths where the expectation would normally be
that they are resolved relative to the configuration file. This
changeset implements config-relative trace file lookups. The old
behavior is kept as a fallback for configs that expect that behavior.
Change-Id: I1bda4e16725842666ffc37dcb6838c23a6ff138c
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Add helper functions to dump the guest kernel's dmesg buffer to a text
file in m5out. This functionality is split into two parts. First, a
dmesg dump function that can be used in other places:
void Linux::dumpDmesg(ThreadContext *, std::ostream &)
This function is used to implement two PCEvents: DmesgDumpEvent and
KernelPanic event. The only difference between the two is that the
latter produces a gem5 panic instead of a warning in addition to
dumping the kernel log.
Change-Id: I6d2af1d666ace57124089648ea906f6c787ac63c
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Gabor Dozsa <gabor.dozsa@arm.com>
The SymbolTable class currently assumes that at most one symbol can
point to a given address. If multiple symbols point to the same
address, only the first one gets added to the internal symbol table
since there is already a match in the address table.
This changeset converts the address table from a map into a multimap
to be able to handle cases where an address maps to multiple
symbols. Additionally, the insert method is changed to not fail if
there is a match in the address table.
Change-Id: I6b4f1d5560c21e49a4af33220efb2a8302961768
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
Reviewed-by: Gabor Dozsa <gabor.dozsa@arm.com>
There is a mismatch between DataType and SrcDataType in constructing
Atomic ST instruction. The mismatch causes atomic_store and
atomic_store_explicit function to store incorrect value in memory.
Eliminate the VSZ constant that defined the Wavefront size (in numbers of work
items); replaced it with a parameter in the GPU.py configuration script.
Changed all data structures dependent on the Wavefront size to be dynamically
sized. Legal values of Wavefront size are 16, 32, 64 for now and checked at
initialization time.
This patch fixes the order that packets gets pushed into the output fifo
of etherswitch. If two packets arrive at the same tick to the etherswitch,
we sort and push them based on their source port id.
In dist-gem5 simulations, if there is no ordering inforced while two
packets arrive at the same tick, it can lead to non-deterministic simulations
Committed by Jason Lowe-Power <power.jg@gmail.com>
mathexpr.hh uses std::function<> but was not including the appropriate
header, which resulted in an error
build/ARM/sim/mathexpr.hh:51:18: error: 'function' in namespace 'std'
does not name a template type
typedef std::function<double(std::string)> EvalCallback;
This commit adds the missing include.
Change-Id: I6c01d77d4354c6de838538f137a38f75f9866166
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
A lot of objects seem to request no-op power transitions which
currently results in large amounts of warnings. These warnings are
benign and risk hiding more serious warnings. Make the warning a
warn_once to prevent console flooding.
Change-Id: I86c74b4224b167f14469250ef86ab69fde7a227e
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
This patch makes thermal models use the power figures
calculated by power models. Since there is a circular
dependency between power and thermal (and thermal was
pushed before) this patch closes that loop.
Change-Id: I8bd5acf6a5026fdbbcfac47e33d27397f24a6f7d
Reviewed-by: Stephan Diestelhorst <stephan.diestelhorst@arm.com>
This patch adds some basic support for power models in gem5.
The power interface is defined so it can interact with thermal
models as well. It implements a simple power evaluator that
can be used for simple power models that express power in the
form of a math expression. These expressions can use stats
within the same SimObject (or down its hierarchy) and some
magic variables such as "temp" for temperature.
In future patches we will extend this functionality to allow
slightly more complex expressions.
The model allows it to be extended to use other kinds of models.
Change-Id: I76752f9638b6815e229fd74cdcb7721a305cbc4b
Add functionality to the BaseCPU that will put the entire CPU
into a low-power idle state whenever all threads in it are idle.
Change-Id: I984d1656eb0a4863c87ceacd773d2d10de5cfd2b
The power stats are currently very noisy even if no power model has
been loaded. Silence stats that are either zero or nan.
Change-Id: I7d0220c2fcf01131084a219228f140cfaddaf95b
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Add 4 power states to the ClockedObject, provides necessary access
functions to check and update the power state. Default power state
is UNDEFINED, it is responsibility of the respective simulation
model to provide the startup state and any other logic for state
change. Add number of transition stat. Add distribution of time
spent in clock gated state. Add power state residency stat. Add
dump call back function to allow stats update of distribution
and residency stats.
Change-Id: Id086090a2ed720c9fcb37812a3c98f0f724907c6
Fixing an issue with regStats not calling the parent class method
for most SimObjects in Gem5. This causes issues if one adds new
stats in the base class (since they are never initialized properly!).
Change-Id: Iebc5aa66f58816ef4295dc8e48a357558d76a77c
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
We want to extend the stats of objects hierarchically and thus it is necessary
to register the statistics of the base-class(es), as well. For now, these are
empty, but generic stats will be added there.
Patch originally provided by Akash Bagdia at ARM Ltd.
This implements SwapReq for Ruby memory.
A SwapReq should be treated like a write, except that the response
packet contains the overwritten data.
Note that, in particular, the conditional checking for isStore/isLoad
needs to be reversed, as a SwapReq is both.
In particular, when EL0 is in AArch32 but EL1 is AArch64, AArch64
memory translation must be used. This is essential for typical
AArch64/32 interworking use cases.
The ERET instruction doesn't set PSTATE correctly in some cases
(particularly when returning to aarch32 code). Among other things,
this breaks EL0 thumb code when using a 64-bit kernel. This changeset
updates the ERET implementation to match the ARM ARM.
Change-Id: I408e7c69a23cce437859313dfe84e68744b07c98
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Nathanael Premillieu <nathanael.premillieu@arm.com>
The current implementation of aarch32 FP/SIMD in gem5 assumes that EL1
and higher are all 32-bit. This breaks interprocessing since an
aarch64 EL1 uses different enable/disable bits. This change updates
the permission checks to according to what is prescribed by the ARM
ARM.
Change-Id: Icdcef31b00644cfeebec00216b3993aa1de12b88
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Mitch Hayenga <mitch.hayenga@arm.com>
Reviewed-by: Nathanael Premillieu <nathanael.premillieu@arm.com>
LPAE has been tested with Linux 4.4 and seems to work just fine. Let's
enable it by default.
Change-Id: Id88c6e3c91ae9c353279d42f2aa1f8a78485bd32
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Gabor Dozsa <gabor.dozsa@arm.com>
According to the ARM ARM (see AArch32.TranslateAddress in the
pseudocode library), the TLB should be operating in aarch64 mode if
the EL0 is aarch32 and EL1 is aarch64. This is currently not the case
in gem5, which breaks 64/32 interprocessing. Update the check to match
the reference manual.
Change-Id: I6f1444d57c0e2eb5f8880f513f33a9197b7cb2ce
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Mitch Hayenga <mitch.hayenga@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
After reaching consensus on the mailing list, this patch officially
makes gcc 4.8 the minimum.
A few checks in the SConstruct are cleaned up as a result. This patch
also adds "-fno-omit-frame-pointer" when using ASAN (which is part of
the gcc/clang recommended flags).
We currently check the current state instead of the state of the
target EL when determining how we report a fault. This breaks
interprocessing since EL0 in aarch32 would report its fault status
using the aarch32 registers even if EL1 is in aarch64. Fix this to
report the fault using the format of the target EL.
Change-Id: Ic080267ac210783d1e01c722a4ddaa687dce280e
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Mitch Hayenga <mitch.hayenga@arm.com>
The TLB currently assumes that the pxn bit in an LPAE page descriptor
disables execution from unprivileged mode. However, according to the
architecture manual, this bit should disable execution from privileged
modes. Update the TLB implementation to reflect this behavior.
Change-Id: I7f1bb232d7a94a93fd601a9230223195ac952947
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
A lot of code assumes that it is possible to test what the highest EL
is and if it is 64 bit. These calls currently don't work in SE mode
since they rely on an instance of an ArmSystem.
Change-Id: I0d1f261926a66ce3dc4fa116845ffb2a081446f2
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Nathanael Premillieu <nathanael.premillieu@arm.com>
This patch fixes a memory leak where deferred snoop packets never got
deallocated. On the call to MSHR::handleSnoop these snoops were
treated as if a response will be sent, as the MSHR was
pendingModified. Consequently, a copy of the packet was created and
added to the MSHR targets. However, an preceeding target to the same
MSHR, originally from a CPU, was serviced before the snoop, and caused
the block to be invalidated. This happens for ReadExReq and
UpgradeReq.
Note that the original snoop will receive a response, just not from
the cache in question, but instead from the cache upstream that issued
the ReadExReq or UpgradeReq.
Change-Id: I4ac012fbc8a46cf693ca390fe9476105d444e6f4
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Make it possible to disable gem5 gic extensions by setting the
gem5_extensions param to False from Python.
Change-Id: Icb255105925ef49891d69cc9fe5cc55578ca066d
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Geoffrey Blake <geoffrey.blake@arm.com>
This patch adds a progress check to the TrafficGen so that it is
easier to detect deadlock scenarios where the generator gets stuck
waiting for a retry, and makes no further progress.
Change-Id: Ifb8779ad0939f52c0518d0e867bac73f99b82e2b
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Radhika Jagtap <radhika.jagtap@arm.com>
This patch changes the flow control for HSHR::handleSnoop to ensure
that we only set cacheResponding on the snoop packet if we are
actually responding. This avoids situations where a responder is
stalling indefinitely on a response that never arrives.
Change-Id: I691dd01755b614b30203581aa74fc743b350eacc
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
This patch fixes the type of the unique_ptr instances, to ensure that
the data that is allocated with new[] is also deleted with
delete[]. The issue was highlighted by ASAN.
Change-Id: I2c5510424959d862a9954d83e728d901bb18d309
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-by: Stephan Diestelhorst <stephan.diestelhorst@arm.com>
This patch fixes an issue identified by ASAN where the Neon64Load
operation assumes the packet always contains 16 bytes.
Change-Id: If24a7e461d60cb80970dfbe61d923d7d56926698
Reviewed-by: Giacomo Gabrielli <giacomo.gabrielli@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
As identified by clang-3.8, there was a superfluous loop increment in
the flash device which is now removed.
Change-Id: If46a1c4f72d3d4c9f219124030894ca433c790af
Reviewed-by: Rene De Jong <rene.dejong@arm.com>
According to the Intel Multi Processor Specification rev 1.4 (-006) (*),
section 4.3.2 Bus Entries, Bus type strings are >>6-character ASCII
(blank-filled) strings<<.
This patch properly pads the entries with the missing spaces at the end.
(*) http://www.intel.com/design/pentium/datashts/24201606.pdf
Committed by Jason Lowe-Power <power.jg@gmail.com>
Given we do not simulate a FIFO currently there are only two states
we can be in upon read: empty or full. Properly signal the latter.
Add and sort constants for states in the header file.
Committed by Jason Lowe-Power <power.jg@gmail.com>
Registers are 0x10 and not 0x8 apart. The latter leads to invalid
calculations of index in array which in turn means that we will not
find the interrupt we were looking (been notified) for in the OS.
Committed by Jason Lowe-Power <power.jg@gmail.com>
VirtIO spec < 1.0 demands IOREG to be used on PCI and not memory mapped.
Set the correct bit on the PCI address accordingly.
Committed by Jason Lowe-Power <power.jg@gmail.com>
the n_reg field in the GPUDynInst is not currently set in the constructor.
if it is not set externally, there are assertion failures that may occur
if the random value it gets is just right. here we set it to 0 by default.
The GICv2 has a new and slightly more consistent register
naming. Update gem5's GIC register names to match the new
documentation.
Change-Id: I8ef114eee8a95bf0b88b37c18a18e137be78675a
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
The LinuxArmSystem class normally provides support for panicing gem5
if the simulated kernel panics. When this is turned off (default),
gem5 uses a BreakPCEvent to provide a debugger hook into the simulator
when the kernel crashes. This hook unconditionally kills gem5 with a
SIGTRAP unless gem5 is compiled in fast mode. This is undesirable
since the panic_on_panic param already provides similar functionality.
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Add support for overriding the number of interrupt lines in the ARM
KvmGic.
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Radhika Jagtap <radhika.jagtap@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Factor out the kernel device wrapper from the KvmGIC and put it in a
separate class. This will simplify a future kernel/gem5 hybrid GIC.
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Radhika Jagtap <radhika.jagtap@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
The Terminal device currently uses the peek functionality in gem5's
circular buffer implementation to send existing buffered content on
the terminal when a new client attaches. This functionallity is
however not implemented correctly and re-sends the same block multiple
time.
Add the required functionality to peek with an offset into the
circular buffer and change the Terminal::accept() implementation to
send the buffered contents.
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Radhika Jagtap <radhika.jagtap@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Allow usage of packet class in ruby for convenience purposes. This may be
used to access members of the packet/request class (e.g., via helper
functions) and/or push protocol specific information to the packets
SenderState without needing to modify SLICC types and protocols in multiple
locations.
This patch removes the write-queue entry tracking previously used for
uncacheable writes. The write-queue entry is now deallocated as soon
as the packet is sent. As a result we also forego the stats for
uncacheable writes. Additionally, there is no longer a need to attach
the write-queue entry to the packet.
Ruby's controller block_on behavior aimed to block MessageBuffer requests into
SLICC controllers when a Locked_RMW was in flight. Unfortunately, this
functionality only partially works: When non-Locked_RMW memory accesses are
issued to the sequencer to an address with an in-flight Locked_RMW, the
sequencer may pass those accesses through to the controller. At the controller,
a number of incorrect activities can occur depending on the protocol. In
MOESI_hammer, for example, an intermediate IFETCH will cause an L1D to L2
transfer, which cannot be serviced, because the block_on functionality blocks
the trigger queue, resulting in a deadlock. Further, if an intermediate store
arrives (e.g. from a separate SMT thread), the sequencer allows the request
through to the controller, and the atomicity of the Locked_RMW may be broken.
To avoid these problems, disallow the Sequencer from passing any memory
accesses to the controller besides Locked_RMW_Write when a Locked_RMW is in-
flight.
Remve the assertion that we always need to add a delta larger than
zero as that does not seem to be true when we hit it in the
'PMU reset cycle counter to zero' case.
Committed by Jason Lowe-Power <power.jg@gmail.com>
A few warnings (and thus errors) pop up after being added to -Wall:
1. -Wmisleading-indentation
In the auto-generated code there were instances of if/else blocks that
were not indented to gcc's liking. This is addressed by adding braces.
2. -Wshift-negative-value
gcc is clever enougn to consider ~0 a negative constant, and
rightfully complains. This is addressed by using mask() which
explicitly casts to unsigned before shifting.
That is all. Porting done.
Queued prefetcher entries now count with a priority field. The idea is to
add packets ordered by priority and then by age.
For the existing algorithms in which priority doesn't make sense, it is set
to 0 for all deferred packets in the queue.
Some common functionality added to the base prefetcher, mainly dealing with
extracting the block address, page address, block index inside the page and
some other information that can be inferred from the block address. This is
used for some prefetching algorithms, and having the methods in the base,
as well as the block size and other information is the sensible way.
In general, the ThreadID parameter is unnecessary in the memory system
as the ContextID is what is used for the purposes of locks/wakeups.
Since we allocate sequential ContextIDs for each thread on MT-enabled
CPUs, ThreadID is unnecessary as the CPUs can identify the requesting
thread through sideband info (SenderState / LSQ entries) or ContextID
offset from the base ContextID for a cpu.
This is a re-spin of 20264eb after the revert (bd1c6789) and includes
some fixes of that commit.
This patch adds a configurable indirect branch predictor that can be indexed
by a combination of GHR and path history hashes. Implements the functionality
described in:
"Target prediction for indirect jumps" by Chang, Hao, and Patt
http://dl.acm.org/citation.cfm?id=264209
This is a re-spin of fb9d142 after the revert (bd1c6789).
The extant BTB code doesn't hash on the thread id but does check the
thread id for 'btb hits'. This results in 1-thread of a multi-threaded
workload taking a BTB entry, and all other threads missing for the same branch
missing.
This changeset updates the dot output to bail out if it is unable to
resolve the voltage or clock domains (which will cause it to raise an
AttributeError). Additionally, the DVFS dot output is disabled by
default for speed purposes.
Minor fixup for 0aeca8f.
The following patches had unexpected interactions with the current
upstream code and have been reverted for now:
e07fd01651f3: power: Add support for power models
831c7f2f9e39: power: Low-power idle power state for idle CPUs
4f749e00b667: power: Add power states to ClockedObject
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
--HG--
extra : amend_source : 0b6fb073c6bbc24be533ec431eb51fbf1b269508
In general, the ThreadID parameter is unnecessary in the memory system
as the ContextID is what is used for the purposes of locks/wakeups.
Since we allocate sequential ContextIDs for each thread on MT-enabled
CPUs, ThreadID is unnecessary as the CPUs can identify the requesting
thread through sideband info (SenderState / LSQ entries) or ContextID
offset from the base ContextID for a cpu.
This patch adds a configurable indirect branch predictor that can be indexed
by a combination of GHR and path history hashes. Implements the functionality
described in:
"Target prediction for indirect jumps" by Chang, Hao, and Patt
http://dl.acm.org/citation.cfm?id=264209
The extant BTB code doesn't hash on the thread id but does check the
thread id for 'btb hits'. This results in 1-thread of a multi-threaded
workload taking a BTB entry, and all other threads missing for the same branch
missing.
This patch adds some basic support for power models in gem5.
The power interface is defined so it can interact with thermal
models as well. It implements a simple power evaluator that
can be used for simple power models that express power in the
form of a math expression. These expressions can use stats
within the same SimObject (or down its hierarchy) and some
magic variables such as "temp" for temperature.
In future patches we will extend this functionality to allow
slightly more complex expressions.
The model allows it to be extended to use other kinds of models.
Finally, the thermal model is updated to use the power usage as input.
Add 4 power states to the ClockedObject, provides necessary access functions
to check and update the power state. Default power state is UNDEFINED, it is
responsibility of the respective simulation model to provide the startup state
and any other logic for state change.
Add number of transition stat.
Add distribution of time spent in clock gated state.
Add power state residency stat.
Add dump call back function to allow stats update of distribution and residency
stats.
This patch enables Linux to read the temperature using hwmon infrastructure.
In order to use this in your gem5 you need to compile the kernel using the
following configs:
CONFIG_HWMON=y
CONFIG_SENSORS_VEXPRESS=y
And a proper dts file (containing an entry such as):
dcc {
compatible = "arm,vexpress,config-bus";
arm,vexpress,config-bridge = <&v2m_sysreg>;
temp@0 {
compatible = "arm,vexpress-temp";
arm,vexpress-sysreg,func = <4 0>;
label = "DCC";
};
};
This patch adds basic thermal support to gem5. It models energy dissipation
through a circuital equivalent, which allows us to use RC networks.
This lays down the basic infrastructure to do so, but it does not "work" due
to the lack of power models. For now some hardcoded number is used as a PoC.
The solver is embedded in the patch.
This patch adds a secondary dot output file which shows the DVFS domains. This
has been done separately for now to avoid cluttering the already existing
diagram. Due to the way that the clock domains are assigned to components in
gem5, this output must be generated after the C++ objects have been
instantiated. This further motivates the need to generate this file separately
to the current dot output, and not to replace it entirely.
This patch adds some additional information when draining the system which
allows the user to debug which SimObject(s) in the system is failing to drain.
Only enabled for the builds with tracing enabled and is subject to the Drain
debug flag being set at runtime.
This patch addresses an issue with the unserialization of clock
domains. Previously, the previous performance level was not restored
due to a bug in the code, which detected the post-unserialize update
as superfluous. This patch splits the setting of the clock domain into
two parts. The original interface of perfLevel is retained, but the
actual update takes place in signalPerfLevelUpdate, which is private
to the class. The perfLevel method checks that if the new performance
level is different to the previous performance level, and will only
call signalPerfLevelUpdate if there is a change. Therefore, the
performance level is only updated, and voltage domains notified, if
there is an actual change. The split functionality allows
signalPerfLevelUpdate to be called by startup() to explicitly force an
update post unserialization.
This patch adds the ability for the simulator to query the number of
instructions a CPU has executed so far per hw-thread. This can be used
to enable more flexible periodic events such as taking checkpoints
starting 1s into simulation and X instructions thereafter.
The openFlagTable and mmapFlagTables for emulated Linux
platforms are basically identical, but are specified
repetitively for every platform. Use a common file
that gets included for each platform so that we only
have one copy, making them more consistent and simplifying
changes (like adding #ifdefs).
In the process, made some minor fixes that slipped through
due to previous inconsistencies, and added more #ifdefs
to try to fix building on alternative hosts.
The style refactor change (style: Refactor the style checker as a
Python package) moved region.py from src/python/m5/util/ to
util/style/. The SConscript update accidentally got lost in that
commit. This commit removes region.py from src/python/SConscript.
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
--HG--
extra : amend_source : f69b75bf636dd4a4232af3e10c29f7eaa4d59dc8
Refactor the style checker into a Python module that can be reused by
command line tools that integrate with git. In particular:
* Create a style package in util
* Move style validators from style.py to the style/validators.py.
* Move style verifiers from style.py to the style/verifiers.py.
* Move utility functions (sort_includes, region handling,
file_types) into the style package
* Move generic code from style.py to style/style.py.
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-by: Steve Reinhardt <steve.reinhardt@amd.com>
--HG--
rename : util/style.py => util/hgstyle.py
rename : util/sort_includes.py => util/style/sort_includes.py
extra : rebase_source : ad6cf9b9a18c48350dfc7b7c77bea6c5344fb53c
This changeset adds an option to force the kvm-based CPUs to always
synchronize the gem5 thread context representation on entry/exit into
the kernel. This is very useful for debugging. Unfortunately, it is
also the only way to get reliable register contents when using remote
gdb functionality. The long-term solution for the latter would be to
implement a kvm-specific thread context.
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-by: Alexandru Dutu <alexandru.dutu@amd.com>
Refactor the TLB and page table walker test interface to use a dynamic
registration mechanism. Instead of patching a couple of empty methods
to wire up a TLB tester, this change allows such testers to register
themselves using the setTestInterface() method.
Libraries are loaded into the process address space using the
mmap system call. Conveniently, this happens to be a good
time to update the process symbol table with the library's
incoming symbols so we handle the table update from within the
system call.
This works just like an application's normal symbols. The only
difference between a dynamic library and a main executable is
when the symbol table update occurs. The symbol table update for
an executable happens at program load time and is finished before
the process ever begins executing. Since dynamic linking happens
at runtime, the symbol loading happens after the library is
first loaded into the process address space. The library binary
is examined at this time for a symbol section and that section
is parsed for symbol types with specific bindings (global,
local, weak). Subsequently, these symbols are added to the table
and are available for use by gem5 for things like trace
generation.
Checkpointing should work just as it did previously. The address
space (and therefore the library) will be recorded and the symbol
table will be entirely recorded. (It's not possible to do anything
clever like checkpoint a program and then load the program back
with different libraries with LD_LIBRARY_PATH, because the
library becomes part of the address space after being loaded.)
The mmapGrowsDown() method was a static method on the OperatingSystem
class (and derived classes), which worked OK for the templated syscall
emulation methods, but made it hard to access elsewhere. This patch
moves the method to be a virtual function on the LiveProcess method,
where it can be overridden for specific platforms (for now, Alpha).
This patch also changes the value of mmapGrowsDown() from being false
by default and true only on X86Linux32 to being true by default and
false only on Alpha, which seems closer to reality (though in reality
most people use ASLR and this doesn't really matter anymore).
In the process, also got rid of the unused mmap_start field on
LiveProcess and OperatingSystem mmapGrowsUp variable.
For O3, which has a stat that counts reg reads, there is an additional
reg read per mmap() call since there's an arg we no longer ignore.
Otherwise, stats should not be affected.
The structure definition only had the open system call flag set in mind when
it was named, so we rename it here with the intention of using it to define
additional tables to translate flags for other system calls in the future.
Breaks the debug output from system calls into two levels: Base and Verbose.
A macro is added specifically for system calls which allows developers to
easily add new debug messages in a consistent manner. The macro also contains
a field to print thread IDs along with the CPU ID.
The cache queue reserve is there as an overflow to give us enough
headroom based on when we block the cache, and how many transactions
we may already have accepted before actually blocking. The previous
values were probably chosen to be "big enough", when we actually know
that we check the MSHRs after every single allocation, and for the
write buffers we know that we implicitly may need one entry for every
outstanding MSHR.
* * *
mem: Adjust cache queue reserve to more conservative values
The cache queue reserve is there as an overflow to give us enough
headroom based on when we block the cache, and how many transactions
we may already have accepted before actually blocking. The previous
values were probably chosen to be "big enough", when we actually know
that we check the MSHRs after every single allocation, and for the
write buffers we know that we implicitly may need one entry for every
outstanding MSHR.
This patch breaks out the cache write buffer into a separate class,
without affecting any stats. The goal of the patch is to avoid
encumbering the much-simpler write queue with the complex MSHR
handling. In a follow on patch this simplification allows us to
implement write combining.
The WriteQueue gets its own class, but shares a common ancestor, the
generic Queue, with the MSHRQueue.
It's apparently not widely known that our scons scripts allow you to
put the build directory wherever you want; not only does it not have
to be immediately under the root of your repo, it doesn't even have
to be underneath the root at all. (For example, sometimes it's useful
to build on a local disk if your repo is on a slow NFS mount.)
I point this out because this functionality has been broken for close
to two years but no one seems to have noticed yet. This patch fixes
an assumption that crept in in changeset be0e1724eb39 (May 09 2014)
that the build dir would be immediately under the top level of the
repo, preventing builds anywhere else.
Adding voltage function which returns the current voltage
for a given clocked object. It's handy for power models and
similar stuff that need to retrieve voltage. Function
frequency() is already there, so I see no reason for not having
this one too.
fu_pool and inst_queue were using -1 for "no such FU" and -2 for "all those
FUs are busy at the moment" when requesting for a FU and replying. This
patch introduces new constants NoCapableFU and NoFreeFU respectively.
In addition, the condition (idx == -2 || idx != -1) is equivalent to
(idx != -1), so this patch also simplifies that.
--HG--
extra : rebase_source : 4833717b9d1e09d7594d1f34f882e13fc4b86846
We can't/shouldn't use KVM after a fork since the child and parent
probably point to the same VM. Knowing the exact effects of this is
hard, but they are likely to be messy. We also disconnect the
performance counters attached to the guest. This works around what
seems to be a kernel bug where spurious SIGIOs get delivered to the
forked child process.
Signed-off-by: Andreas Sandberg <andreas@sandberg.pp.se>
[sascha.bischoff@arm.com: Rebased patches onto a newer gem5 version]
Signed-off-by: Sascha Bischoff <sascha.bischoff@arm.com>
[andreas.sandberg@arm.com: Fatal if entering KVM in child process ]
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
This changeset adds forking capabilities to the gem5 python scripts. A fork
method is added to simulate.py. This method is responsible for forking the
simulator itself, and will direct all output files to a new output directory
based on the fork sequence number. The default name of the output directory is
the same as the parent with the suffix ".fN" added where N is the fork sequence
number. The fork method provides the option to specify if the system should be
drained prior to forking, or not. By default the system is drained to ensure
that there are no in-flight transactions.
When forking the simulator, the fork method returns the PID of the child
process, or returns 0 if running in the child. This is in line with the standard
Python forking interface.
Signed-off-by: Andreas Sandberg <andreas@sandberg.pp.se>
[sascha.bischoff@arm.com: Rebased patches onto a newer gem5 version]
Signed-off-by: Sascha Bischoff <sascha.bischoff@arm.com>
[andreas.sandberg@arm.com: Updated to comply with modern draining semantics ]
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
This changeset adds support for notifying the disk images that the simulator has
been forked. We need to disable the saving of the CoW disk image from the child
process, and we need to make sure that systems which use a raw disk image are
not allowed to fork to avoid two or more gem5 processes writing to the same disk
image.
Signed-off-by: Andreas Sandberg <andreas@sandberg.pp.se>
[sascha.bischoff@arm.com: Rebased patches onto a newer gem5 version]
Signed-off-by: Sascha Bischoff <sascha.bischoff@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
When forking a gem5 process, some objects need to clean up resources
(mainly file descriptions) shared between the child and the parent of
the fork. This changeset adds the notifyFork() method to Drainable,
which is called in the child process.
Signed-off-by: Andreas Sandberg <andreas@sandberg.pp.se>
[sascha.bischoff@arm.com: Rebased patches onto a newer gem5 version]
Signed-off-by: Sascha Bischoff <sascha.bischoff@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
This changeset adds support for changing the simulator output
directory. This can be useful when the simulation goes through several
stages (e.g., a warming phase, a simulation phase, and a verification
phase) since it allows the output from each stage to be located in a
different directory. Relocation is done by calling core.setOutputDir()
from Python or simout.setOutputDirectory() from C++.
This change affects several parts of the design of the gem5's output
subsystem. First, files returned by an OutputDirectory instance (e.g.,
simout) are of the type OutputStream instead of a std::ostream. This
allows us to do some more book keeping and control re-opening of files
when the output directory is changed. Second, new subdirectories are
OutputDirectory instances, which should be used to create files in
that sub-directory.
Signed-off-by: Andreas Sandberg <andreas@sandberg.pp.se>
[sascha.bischoff@arm.com: Rebased patches onto a newer gem5 version]
Signed-off-by: Sascha Bischoff <sascha.bischoff@arm.com>
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
This patch adds assertions that enforce that only invalidating snoops
will ever reach into the logic that tracks in-order load completion and
also invalidation of LL/SC (and MONITOR / MWAIT) monitors. Also adds
some comments to MSHR::replaceUpgrades().
Writes to locked memory addresses (LLSC) did not wake up the locking
CPU. This can lead to deadlocks on multi-core runs. In AtomicSimpleCPU,
recvAtomicSnoop was checking if the incoming packet was an invalidation
(isInvalidate) and only then handled a locked snoop. But, writes are
seen instead of invalidates when running without caches (fast-forward
configurations). As as simple fix, now handleLockedSnoop is also called
even if the incoming snoop packet are from writes.
Properly done for the ERET instruction in v8, but not for v7.
Many control register changes are only visible after explicit
instruction synchronization barriers or exception entry/exit.
This means mode changing instructions should squash any
younger in-flight speculative instructions.
This patch fixes an issue where an InvalidationReq only traversed one
level of the cache hierarchy, and was subsequently turned into a
ReadExReq due to it needing writable, and the command not being
checked for explicitly.
Bug fix for check on protobuf file frequency being different than
global frequency.
The ASCII encoder script is also fixed, and the example trace used in
the regressions is updated.
Add a callback handler for the NoMali reset callback. This callback is
called whenever the GPU is reset using the register interface or the
NoMali API. The callback can be used to override ID registers using
the raw register API.
Refactor and cleanup the NoMaliGpu class:
* Use a std::map instead of a switch block to map the parameter enum
describing the GPU type to a NoMali type.
* Remove redundant NoMali handle from the interrupt callback.
* Make callbacks and API wrappers protected instead of private to
enable future extensions.
* Wrap remaining NoMali API calls.
Both Memory Fence is now flagged as Global Memory only to avoid resource
oversubscribing.
Flat instructions now check for Shared Memory resource busy to avoid
oversubscribing resources.
All WaitClass resources now use cycles (not ticks) to register the number
of pipe stages between Scoreboard and Execute to be consistent with
instruction scheduling logic which always used clock cycles.
This patch essentially rolls back 10518:30e3715c9405 to make RubyPort the
parent class of DMASequencer. It removes redundant code and restores some
features which were lost when directly inheriting from MemObject. For
example,
DMASequencer can now communicate to other devices using PIO, which is useful
for memmory-mapped communication between multiple DMADevices.
This patch adds a --debug-end flag to main.py so that debug output can be
stoped at a specified tick, while allowing the simulation to continue. It is
useful in situations where you would like to produce a trace for a region of
interest while still collecting stats for the entire run. This is in contrast
to the currently existing --debug-break flag, which terminates the simulation
at the tick.
Avoid being overly conservative in clearing load locks in the cache,
and allow writes to the line if they are from the same context. This
is in line with ALPHA and ARM.
This patch introduces the ability of making the coherent crossbar the
point of coherency. If so, the crossbar does not forward packets where
a cache with ownership has already committed to responding, and also
does not forward any coherency-related packets that are not intended
for a downstream memory controller. Thus, invalidations and upgrades
are turned around in the crossbar, and the memory controller only sees
normal reads and writes.
In addition this patch moves the express snoop promotion of a packet
to the crossbar, thus allowing the downstream cache to check the
express snoop flag (as it should) for bypassing any blocking, rather
than relying on whether a cache is responding or not.
Adopt the same flow as in timing mode, where the caches on the path to
memory get to keep the line (if present), and we use the
responderHadWritable flag to determine if we need to forward the
(invalidating) packet or not.
This patch unifies the snoop handling in case of hitting writebacks
with how we handle snoops hitting in the tags. As a result, we end up
using the same optimisation as the normal snoops, where we inform the
downstream cache if we encounter a line in Modified (writable and
dirty) state, which enables us to avoid sending out express snoops to
invalidate any Shared copies of the line. A few regressions
consequently change, as some transactions are sunk higher up in the
cache hierarchy.
This patch changes how the cache determines if snoops should be
forwarded from the memory side to the CPU side. Instead of having a
parameter, the cache now looks at the port connected on the CPU side,
and if it is a snooping port, then snoops are forwarded. Less error
prone, and less parameters to worry about.
The patch also tidies up the CPU classes to ensure that their I-side
port is not snooping by removing overrides to the snoop request
handler, such that snoop requests will panic via the default
MasterPort implement
Due to insufficient build deps, the checkpoint tags might not get
updated; this commit solves this. Due to the uncommon nature of the
build target, regenerating tags.cc is a fairly clean solution. Since
SCons hashes file contents, it won't recompile anything unless a new
checkpoint upgrader is actually added.
--HG--
extra : amend_source : ed3879da7668554693f697076deaf5029cc9b954
The previous implementation did a pair of nested RMW operations,
which isn't compatible with the way that locked RMW operations are
implemented in the cache models. It was convenient though in that
it didn't require any new micro-ops, and supported cmpxchg16b using
64-bit memory ops. It also worked in AtomicSimpleCPU where
atomicity was guaranteed by the core and not by the memory system.
It did not work with timing CPU models though.
This new implementation defines new 'split' load and store micro-ops
which allow a single memory operation to use a pair of registers as
the source or destination, then uses a single ldsplit/stsplit RMW
pair to implement cmpxchg. This patch requires support for 128-bit
memory accesses in the ISA (added via a separate patch) to support
cmpxchg16b.
Although the cache models support wider accesses, the ISA descriptions
assume that (for the most part) memory operands are integer types,
which makes it difficult to define instructions that do memory accesses
larger than 64 bits.
This patch adds some generic support for memory operands that are arrays
of uint64_t, and specifically a 'u2qw' operand type for x86 that is an
array of 2 uint64_ts (128 bits). This support is unused at this point,
but will be needed shortly for cmpxchg16b. Ideally the 128-bit SSE
memory accesses will also be rewritten to use this support.
Support for 128-bit accesses could also have been added using the gcc
__int128_t extension, which would have been less disruptive. However,
although clang also supports __int128_t, it's still non-standard.
Also, more importantly, this approach creates a path to defining
256- and 512-byte operands as well, which will be useful for eventual
AVX support.
MemOperand variables were being initialized to 0
"to avoid 'uninitialized variable' errors" but these
no longer seem to be a problem (with the exception of
one use case in POWER that is arguably broken and
easily fixed here).
Getting rid of the initialization is necessary to
set up a subsequent patch which extends memory
operands to possibly not be scalars, making the
'= 0' initialization no longer feasible.
Writing 16 bytes from an 8-byte source value is a bad idea.
This doesn't appear to have broken anything, but showed up
as spurious differences when tracediffing runs.
Result of running 'hg m5style --skip-all --fix-control -a' to get
rid of '== true' comparisons, plus trivial manual edits to get
rid of '== false'/'== False' comparisons.
Left a couple of explicit comparisons in where they didn't seem
unreasonable:
invalid boolean comparison in src/arch/mips/interrupts.cc:155
>> DPRINTF(Interrupt, "Interrupts OnCpuTimerINterrupt(tc) == true\n");<<
invalid boolean comparison in src/unittest/unittest.hh:110
>> "EXPECT_FALSE(" #expr ")", (expr) == false)<<
In the process of trying to get rid of an '== false' comparison,
it became apparent that a slightly more involved solution was
needed. Split this out into its own changeset since it's not
a totally trivial local change like the others.
