cpu: Add first-/non-faulting load support to Minor and O3

Some architectures allow masking faults of memory load instructions in
some specific circumstances (e.g. first-faulting and non-faulting
loads in Arm SVE). This patch adds support for such loads in the Minor
and O3 CPU models.

Change-Id: I264a81a078f049127779aa834e89f0e693ba0bea
Signed-off-by: Gabor Dozsa <gabor.dozsa@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19178
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: Fix SMMUv3 CMDQ wrapping

SMMU circular queues have a wrap bit which is used in order to
distinguish between an empty queue and a full queue.

According to SMMUv3 spec:

Each index has a wrap flag, represented by the next higher bit adjacent
to the index value contained in PROD and CONS. This bit must toggle each
time the index wraps off the high end and back onto the low end of the
buffer. It is the responsibility of the owner of each index, producer or
consumer, to toggle this bit when the owner updates the index after
wrapping. It is intended that software reads the register, increments or
wraps the index (toggling wrap when required) and writes back both wrap
and index fields at the same time.

Change-Id: Idfeb397141f3627c2878caaeaa2625fadf671d2a
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Michiel Van Tol <michiel.vantol@arm.com>
Reviewed-by: Adrian Herrera <adrian.herrera@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19311
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Ciro Santilli <ciro.santilli@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>

dev-arm: Polish SMMUv3 CMDQ setup

The patch is aiming to be spec compliant when it comes to setup
the SMMU command queue (while CR0.CMDQEN = 0), in the following ways:

* Writes to CMDQ_CONS (read index) are allowed during initialization
* Writes to CMDQ_BASE (cmdq pointer) are allowed during initialization

According to spec,
If they happen when the command queue is in fuction (CR0.CMDQEN = 1),
behaviour is constrained unpredictable, with the following options

1) The write is ignored
2) The register takes the value and it is unpredictable whether it
affects the SMMU command queue internal state.

In the model/patch we go for option 1.

Change-Id: I1c55bc571a8b3a1c0b0a525e429ab7b1480544ff
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Michiel Van Tol <michiel.vantol@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19633
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: Define enum masks for SMMU_CR0 register

The configuration register is a vital register in the SMMU, and using
enum masks will make the code more readable/understandable

Change-Id: Ia117db56c457fe876ae38be391c386e502f34384
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Michiel Van Tol <michiel.vantol@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19632
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: TnSZ fields need to be cached in SMMUv3::ConfigCache

Otherwise a hit after a table walk will result in a 0 value being
read from the ConfigCache.

Change-Id: I9813998acce44c93c5ce203f252ca80c10ba8f38
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Michiel Van Tol <michiel.vantol@arm.com>
Reviewed-by: Adrian Herrera <adrian.herrera@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19631
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: SMMUv3 Table walks using TnSZ

TnSZ is needed when selecting the starting level of a table
walk, since it directly affects the number of IA bits.
This has been implemented by adding T0SZ and S2T0SZ to the
translation context.
T1SZ is not used at the moment since the current model doesn't
support TTB1.

Change-Id: I75663475c4dc01e5986cd93f8deafcdf7b1ece82
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Michiel Van Tol <michiel.vantol@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19630
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: Use override keyword for SMMUv3 PTOPS

Replacing the "virtual" keyword

Change-Id: I0e7b4b683ea222827a67c3a81f0deea0e906c7e5
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Michiel Van Tol <michiel.vantol@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19629
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: Add 16K granule support to SMMUv3 model

Added the necessary PageTableOps that match the 16K granule
translation regime.

Change-Id: I46ef07939cb4bdc8c0bbbeeeb6a50a9ab0d64de0
Reviewed-by: Michiel Van Tol <michiel.vantol@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19628
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

mem-ruby: Adding a new slicc statement - to not evict locked cachelines

Ruby caches block incoming ports with messages on a locked
address to make sure the line would not be replaced by others.
But they do not check the lock upon capacity/conflict misses.

This change adds a new slicc statement "check_on_cache_probe" which takes
two arguments (mandatoryQueue for the controller, and the line subject
to eviction - i.e. address returned by cacheProbe).
If the line is locked, incoming message is delayed for 1 cycle and the
controller skips this request (i.e. does not trigger an event).

Coherence protocols should be updated accordingly. One use case for MESI
Two Level will be added in a separate change.

Signed-off-by: Pouya Fotouhi <pfotouhi@ucdavis.edu>
Change-Id: I79ca2d45518de7a4e382b520a11f8e221b0cb803
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/16808
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Srikant Bharadwaj <srikant.bharadwaj@amd.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>

arch-x86: Don't free PTW state with inflight requests

If a page table walk is squashed, the walker state is being deleted
in the squash code. If there are in flight requests, the deleted
walker state values may be clobbered, leading to undefined behavior.
This adds a squashed boolean to the walker state which is set if a
walk is squashed while requests are still in flight. When packets
for the in flight request return, we check if the walk was squashed
and return that the walk is complete once the number of in flight
requests reaches zero. The walker state is then freed by the PTW.

Change-Id: I57a64b1548b83a8a9e8441fc9d6f33e9842df2b3
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19568
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-arm: Implement ARMv8.1-HPD, Hierarchical permission disable

According to the armarm:
ARMv8.1-HPD introduces the facility to disable the hierarchical
attributes, APTable, PXNTable, and UXNTable, in the translation tables.
This disable has no effect on the NSTable bit.  This feature is
mandatory in ARMv8.1 implementations.

This feature is added only to the VMSAv8-64 translation regimes. ARMv8.2
extends this to the AArch32 translation regimes, see ARMv8.2-AA32HPD.

The ID_AA64MMFR1_EL1.HPDS field identifies the support for ARMv8.1-HPD.

Change-Id: Ibbf589b82f2c1e4437b43252f8f633e8f6fb0b80
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Ciro Santilli <ciro.santilli@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19610
Tested-by: kokoro <noreply+kokoro@google.com>

arch-arm: Add HPD bit for TCR_EL2/EL3

This is controlling Hierarchical Permissions for EL3 and for EL2 when
HCR_EL2.E2H = 0.

Change-Id: I9d6615ff4980cc56a28dc32088cf1524155f0fa4
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Ciro Santilli <ciro.santilli@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19609
Tested-by: kokoro <noreply+kokoro@google.com>

arch-arm: Clean Fault generation when processing Long Descriptor

A new shared method has been introduced: generateLongDescFault

Change-Id: I7eb6fa1347a6c2cf9cb11fd9f2137d983c4f7a40
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Ciro Santilli <ciro.santilli@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19608
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: clang compatibility fix, added missing overrides

Change-Id: I5ee5ff788570178bb1d68878a26ac9e3ce636d8e
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19588
Reviewed-by: Matteo Andreozzi <matteo.andreozzi@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-arm: Add first-/non-faulting load instructions

First-/non-faulting loads are part of Arm SVE.

Change-Id: I93dfd6d1d74791653927e99098ddb651150a8ef7
Signed-off-by: Gabor Dozsa <gabor.dozsa@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19177
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

sim: Add getter to fault virtual address

Change-Id: Ifd493aee9e78b0b4ddcc71e90f48679543acb861
Signed-off-by: Giacomo Gabrielli <giacomo.gabrielli@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19176
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

mem-cache: Create CPack compressor

Implementation of C-Pack, as described in "C-Pack: A High-
Performance Microprocessor Cache Compression Algorithm".

C-Pack uses pattern matching schemes to detect and compress
frequently appearing data patterns. As in the original paper,
it divides the input in 32-bit words, and uses 6 patterns to
match with its dictionary.

For the patterns, each letter represents a byte: Z is a null
byte, M is a dictionary match, X is a new value. The patterns
are ZZZZ, XXXX, MMMM, MMXX, ZZZX, MMMX.

Change-Id: I2efc9db2c862620dcc1155300e39be558f9017e0
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/11105
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-arm: Use ExceptionLevel type in TlbEntry

Replacing uint8_t with ExceptionLevel type in the arm TlbEntry. The
variable is representing the translation regime it is targeting.

Change-Id: Ifcd6e86c5d73f752e8476a2b7fda9ea74a0c7a3b
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19488
Reviewed-by: Ciro Santilli <ciro.santilli@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-x86: add unconditional tag to calls/returns

The branch predictor checks whether an instruction is unconditional
branch before adding it or checking the RAS. With this change, the
RAS is significantly more effective for short running x86 workloads.

Change-Id: I60af5f2f583b898ad77f79f4b0478d6cda88fc21
Signed-off-by: Hoa Nguyen <hoanguyen@ucdavis.edu>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19448
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Gabe Black <gabeblack@google.com>

dev-arm: Fix SMMUv3 ContextDescriptor pointer shift

The context descriptor pointer in the STE starts at the sixth LSB

Change-Id: Ifa346b350785b788e9d1e093b662cb26433adfb8
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Christo Smallwood <christo.smallwood@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19469
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

cpu: isDrained renamed to isCpuDrained

cpu models inheriting from BaseCPU implement a draining checker called
isDrained. This hides the base Drainable::isDrained method and might
create confusion in the reader.
This patch is renaming it to isCpuDrained in order to avoid any
ambiguity

Change-Id: Ie5221da6a4673432c2403996e42d451cae960bbf
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19468
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

mem: Check response only when needed in CommMonitor

CommMonitor checks pkt->isResponse() for all packets in recvAtomic().
This assertion fails when packets don't need response, such as WritebackDirty.
This change fixes this.

Signed-off-by: Chun-Chen TK Hsu
Change-Id: I168e349e179b14fa5472698d9300478dc89693fb
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19428
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

cpu-o3: Set packet data type for IPR read

This change assigns packet data type to static for IPR read.
Caused by change (e13d6dc9c0d7a4ae0215f1ee6793eb32570c5169),
and has been reported a few times in the mailing list.

Change-Id: I0f02c20a16824e220df876e9e552bbc1c9636f95
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19449
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: A9SCU fixup

Shifting instead of expensive power.

Change-Id: I164933257db125e18721c5b8bcaf9702030ebf40
Signed-off-by: Tiago Mück <tiago.muck@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19408
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

cpu: Added the Multiperspective Perceptron Predictor with TAGE (8KB and 64KB)

Described by the following article:
  Jiménez, D. "Multiperspective perceptron predictor with TAGE."
  Championship Branch Prediction (CBP-5) (2016).

Change-Id: Ica3c121a4c94657d9015573085040e8a1984b069
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19188
Tested-by: kokoro <noreply+kokoro@google.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Ilias Vougioukas <ilias.vougioukas@arm.com>

cpu-o3: Reset fault status for mem access in pushRequest

Reset the fault status always before translation is initiated in
pushRequest() in the LSQ. This avoids the problem when a strictly
ordered load needs to be re-executed multiple times. If the
translation is delayed at one of those attempts then the
internal panicFault (from the previous execution attempt) can get
fired at commit.

Change-Id: I0c22b2f7afd6e2cb00bc359a4a01042efd2d01d2
Signed-off-by: Gabor Dozsa <gabor.dozsa@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19388
Reviewed-by: Ciro Santilli <ciro.santilli@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

mem-ruby: Fix MultiGrainBloomFilter accessing

When accessing the page filter the page hash should be used
instead of the hash of the base filter.

Change-Id: I17b7c64f2a0d654c7d9a77a7bfb435385d81032c
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18739
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>

mem-ruby: Remove Bloom Filters' ruby dependency

Substitute the common ruby header by base's bitfield to
eliminate all ruby dependency in Bloom Filters.

As a side note, BulkBloomFilter now assumes addresses are
64 bit long.

Change-Id: Ibdb1f926ddcc06c848851c1e6a34863541808360
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18738
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>

mem-ruby: Parameterize block size in Bloom Filters

Substitute all occurrences of Ruby's block size by a
Python configurable offset.

Change-Id: If4913e842921447deda943b0482fb0c78a44c275
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18737
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

mem-ruby: Make Bloom Filters SimObjects

Make all bloom filters SimObjects.

Change-Id: I586293cdfb559361cb868b3198368e8b9b193356
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18736
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>

mem-ruby: Generalize use of bloom filters' isSet

In general the corresponding entries of an address are considered
to be set when the sum of all of them reach their maximum value
(i.e., they are all set), so generalize that into the base class.

Change-Id: If50b8c56065ad339b4ff2322ddc3c077a3bfc518
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18735
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>

mem-ruby: Cleanup filters

Renamed member variables to comply with general naming
conventional outside of the ruby folder so that the
filters can be moved out.

Moved code to base to reduce code duplication.

Renamed the private get_index functions to hash, to make their
functionality explicit.

Change-Id: Ic6519cfc5e09ea95bc502a29b27f750f04eda754
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18734
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>

mem-ruby: Fix MultiGrainBloomFilter total count calculation

Previous value was always 0, and was never incrementing. The
total count should take into account the value stored in the
entry.

Change-Id: I93813e3f388198967b30cf11848a8a8c3a7b91f4
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18733
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>

mem-ruby: Remove index based functions in bloom filters

Most of the index based functions were not implemented, and a
user is more likely to be interested in checking the filter
contents based on an address than an index.

As a side effect, the Bulk's hash function became unused, and
according to the paper permute() was doing more than just
permuting, so it was renamed.

Change-Id: I6423a2565a082fee2e7f11fa489a11f253064d99
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18732
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>

mem-ruby: Remove Bloom Filter's print()

Print was unused. As a side effect 'using namespace std' is no
longer needed.

Change-Id: Ief10cba1a11dfdd4edb7464eb9291fc83d6668cd
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18731
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>

mem-ruby: Standardize Bloom Filter deletion support

Standard Bloom Filters do not support element deletion by default,
however some variants do. Allow calling the unset function with
all filters, and do nothing by default.

Change-Id: Icf4b0f8b997c4c70fa714b2576474810275db78b
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18730
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>

mem-ruby: Bloom filters - Remove in/decrement

Increment and decrement were functions created to supply the
different naming convention used by the counting bloom filter.
They were removed, and the set and unset functions were used
in their place instead, as in the other filters.

Change-Id: I45732bdfa3083add0a975f374a0f3560003e9d09
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18729
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>

dev-arm: Use global import path for MemObject

Change-Id: I66e0ca6df689ec6aeb831ef5545e8e5842bb0418
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Ciro Santilli <ciro.santilli@arm.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19348
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

base: Add argument to Coroutine class to not run on creation

In some cases, the point where you create a Coroutine is not the same as
where you want to start running it (and want it to switch back to). This
leads to the unnecessary overhead of switching in and out of the
Coroutine. This change adds an optional boolean argument to the
constructor for the Coroutine class to allow for overriding the default
behavior of running the Coroutine upon creation, which in specific cases
can be used to avoid the unnecessary overhead and improve simulator
performance.

Change-Id: I044698f85e81ee4144208aee30d133bcb462d35d
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18588
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch, arm: Update miscRegs in getTE

Normally, a translation will start via translateTiming/functional
which will check if the miscRegs have been updated and if so,
will update the TLB state accordingly.  However, in a 2 stage
system, if there is a hit in stage 1, the resulting IPA will be
sent to the S2-TLB for translation via a getTE() function call
(via the stage2_lookup object).  This will cause the state of the
S2-TLB to be out of sync.

Change-Id: I117e4032fc76d7d31f4f999887b5573a7e5811e6
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/14995
Tested-by: kokoro <noreply+kokoro@google.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>

dev-arm: Remove un-needed Q_CONS_PROD_MASK macro

Change-Id: I858d7eea088bbdd2dc12123e21e59991c896597f
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Michiel Van Tol <michiel.vantol@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19310
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: drain implementation for SMMUv3

SMMUv3 is drained when (1) no SMMU translations are pending
on any of its slave interfaces and (2) no commands are stored
in the Command Queue waiting to be processed.

Change-Id: I81cef5fd821fa5e509e130af02aece5239493df5
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19309
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: pending SMMU transl update on constructor/destructor

Change-Id: I6f61651123aab129cfbe5a88aa6355cd21544a5e
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19308
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

configs: Fix NULL dram-lowp regressions

The problem arises since there are some scripts (like NULL dram
regressions) which are making use of MemConfig.py without using
Opions.py so they won't have the new enable-dram-powerdown option

Change-Id: Id9769cce2e8a25b57da76f07eeebd279a6e00440
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19268
Reviewed-by: Matthew Poremba <matthew.poremba@amd.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

python: Add support for exporting static class methods from c++

This change adds support for exporting static methods in a c++
SimObject from the coressponsing python wrapper class. This will allow
us to define and use c++ methods without the need to instantiate an
object of the corresponding class.

Change-Id: Iaf24c1aa6f20feb5c91241f46ec8db005a6a0c0c
Signed-off-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Signed-off-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19168
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>

misc: dot_writer fixup

In large configs the tooltip may be greater then the maximum line
size graphviz supports when parsing the dot file (typically 16k).
Adding '/' causes graphviz to break the string in multiple lines
while parsing and works around this limitation.

Change-Id: I16a0030127de4165080de97f5213309eed9fdeca
Signed-off-by: Tiago Mück <tiago.muck@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19208
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

python: Fix AddrRange legacy ParamValue wrapper

This change fixes a bug that would manifest if a user would
instantiate an AddrRange ParamValue using the kwargs 'intlvBits' and
'intlvHighBit' without specifying the optional 'xorHighBit'.

Change-Id: I2091c432234df9cf907d52af6ba7f0cadd8c37a8
Signed-off-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19248
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-arm: Move the memacc_code before op_wb in fp loads

This is trying to fix the bug that arises when a memory exception
is generated during a fp flavoured load (A memory load targeting
a SIMD & FP register).
With the previous template a fault was not stopping the register
value to be modified (wrong)

if (fault == NoFault) {
    fault = readMemAtomic(xc, traceData, EA, Mem, memAccessFlags);
    %(memacc_code)s;
}

if (fault == NoFault) {
    %(op_wb)s;
}

The patch introduces a Load64FpExecute template which is moving the
register write (memacc_code) just before the op_wb

Change-Id: I1c89c525dfa7a4ef489abe0872cd7baacdd6ce3c
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19228
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: Reapply GICv3 changes that were lost during refactoring

The GICv3 code refactoring performed by:

https://gem5-review.googlesource.com/c/public/gem5/+/16484

reverted the following patches

https://gem5-review.googlesource.com/c/public/gem5/+/16544
https://gem5-review.googlesource.com/c/public/gem5/+/16545/3

This commit is reintroducing them

Change-Id: I2c875c11570ed66ec9203449446faca3864c64d6
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Ciro Santilli <ciro.santilli@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19229
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

base: AddrRange does not merge single interleaved ranges

AddrRange does not attempt to merge interleaved address ranges if it
has only one of the ranges.

This is needed to allow XBars to accept request targeting only one
part of a interleaved address range. A use case for this would be
modeling distributed LLCs in which a XBar is used solely to
encapsulate the snoop filter of a single LLC slice.

Change-Id: If71c9cf1444ee11916611afb51eab3a4f1d93985
Signed-off-by: Tiago Muck <tiago.muck@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18788
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

cpu: Additional TrafficGen stats

Additional stats to keep track of read/write latencies and throughput.

Change-Id: I7684cd33cf68fffdef4ca9c3a6db360a0f531c18
Signed-off-by: Tiago Muck <tiago.muck@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18418
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

cpu: Limit TrafficGen outstanding reqs

Parameter to limit the number of requests waiting for a response.

Change-Id: I6cf9e8782a06ae978fb66f7c4278f4c9e9980c79
Signed-off-by: Tiago Muck <tiago.muck@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18417
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

cpu: TrafficGen as BaseCPU

TrafficGen has additional attributes to behave like a BaseCPU. Python
scripts that expect sim. objects derived from BaseCPU can now be used with
TrafficGen without additional modifications.

Change-Id: Iee848b2ba0ac1851c487b1003da9bd96253d291a
Signed-off-by: Tiago Muck <tiago.muck@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18416
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

python: Fix cxx_ini_parse for ScopedEnum

Change-Id: I69534bbbc16b2d24a65fa1dee4926f213c3618ef
Signed-off-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19152
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

configs: Fix read_config to work with new AddrRange serialization

Change-Id: I122c77c34c2f8c75f8b32682be858f651112ce89
Signed-off-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19151
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

configs: Add python3 workarounds in read_config

Change-Id: Ib065f41b87e5ada9535b9c2645067162aa69234b
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19150
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

python: Add binding for the new AddrRange c++ constructor

Change-Id: I5b3fb59a11d8587a753759310dd3b2748ac13a0b
Signed-off-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19132
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

base: Extend unit tests for AddrRange

Change-Id: Ia154c3a17c3c8254a0e3d622568ac34f0d62fc9e
Signed-off-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19131
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

base: Extend AddrRange to support more flexible addressing

Previously an AddrRange could express interleaving using a number of
consecutive bits and in additional optionally a second number of
consecutive bits. The two sets of consecutive bits would be xored and
matched against a value to determine if an address is in the
AddrRange. For example:

sel[0] = a[8] ^ a[12]
sel[1] = a[9] ^ a[13]
where sel == intlvMatch

This change extends AddrRange to allow more flexible interleavings
with an abritary number of set of bits which do not need be
consecutive. For example:

sel[0] = a[8] ^ a[11] ^ a[13]
sel[1] = a[15] ^ a[17] ^ a[19]
where sel == intlvMatch

Change-Id: I42220a6d5011a31f0560535762a25bfc823c3ebb
Signed-off-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19130
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>

base: Fix ctz32 for systems where unsigned int is not 32bit

The implementation of ctz32 uses __builtin_ctz to count the number of
trailing zeros and therefore makes the assumption that an unsigned int
is 32bit. This change checks whether that's the case and if not it
uses __builtin_ctzl instead.

Change-Id: Ic3ed3ada25fd0a93c7eb91d75b954e9924bdbb77
Signed-off-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19129
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

base: Add function to count trailing zeros in a 64-bit integer

Change-Id: Iaad0679b403bc5015ffeacbf7284313e41a36cd0
Signed-off-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19128
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

scons: allow passing arbitrary CCFLAGS and LDFLAGS from the CLI

The flags may be passed as:

scons CCFLAGS_EXTRA='-Wno-error -pedantic' \
      LDFLAGS_EXTRA='-g -g' build/<arch>/gem5.opt

The initial motivation for this commit is to help disable warning that
have become errors while bisecting.

scons orders the flags by Append call order, and ideally these flags
should be added last to override the others, since the last GCC flags
take precedence. However I haven't found a simple way to put them at
the very end.

Change-Id: Ida24dfb9604d88b99f113392ab5e47d578ba7259
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19048
Reviewed-by: Juha Jäykkä <juha.jaykka@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-arm: implement VMINNM scalar thumb

VMINNM was implemented at Iabbbca2932557cf6c98ce36690c385c3ddf39ed8 but
the thumb scalar encoding was missing. This patch implements it.

Change-Id: Ia29ec77dbd82f6be6b3d040a0e737794f52c33bf
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19108
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

base: Provide a getter for Fiber::started boolean variable

This can be used to check if the fiber has started its execution.

Change-Id: Ie9222b8076756363c9f82c1333c76a352bcaf817
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18648
Reviewed-by: Gabe Black <gabeblack@google.com>
Maintainer: Gabe Black <gabeblack@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>

base: Rename TestFiber into SwitchingFiber

This is done since TestFiber is too generic and it is not the only
Fiber's testing subclass in the unit test

Change-Id: Idc386f487091ae9bdadae865090a6719a25583da
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19153
Reviewed-by: Gabe Black <gabeblack@google.com>
Maintainer: Gabe Black <gabeblack@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch-arm: Fix WalkerState,Descriptors default constructor

Those POD strutures are not initializing all members at construction.
This could lead to undefined behaviour

Change-Id: Iaa8afb126382b6bfbef686883a026262f24d5ca1
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Ciro Santilli <ciro.santilli@arm.com>
Reviewed-by: Javier Setoain <javier.setoain@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19149
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

dev-arm: Implement a SMMUv3 model

This is an implementation of the SMMUv3 architecture.

What can it do?
- Single-stage and nested translation with 4k or 64k granule.  16k would
  be straightforward to add.
- Large pages are supported.
- Works with any gem5 device as long as it is issuing packets with a
  valid (Sub)StreamId

What it can't do?
- Fragment stage 1 page when the underlying stage 2 page is smaller.  S1
  page size > S2 page size is not supported
- Invalidations take zero time. This wouldn't be hard to fix.
- Checkpointing is not supported
- Stall/resume for faulting transactions is not supported

Additional contributors:
- Michiel W. van Tol <Michiel.VanTol@arm.com>
- Giacomo Travaglini <giacomo.travaglini@arm.com>

Change-Id: Ibc606fccd9199b2c1ba739c6335c846ffaa4d564
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19008
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

mem: Option to toggle DRAM low-power states

Adding an option to enable DRAM low-power states. The low power
states can have a significant impact on application performance
(sim_ticks) on the order of 2-3x, especially for compute-gpu apps.
The options allows for it to easily be enabled/disabled to compare
performance numbers. The option is disabled by default.

Change-Id: Ib9bddbb792a1a6a4afb5339003472ff8f00a5859
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18548
Reviewed-by: Wendy Elsasser <wendy.elsasser@arm.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

mem-ruby: Enable set size increase

Add NUMBER_BITS_PER_SET environment variable to control
the size of the bitmask in Set.hh (default=64).
Necessary for configs which require >64 instances of a given
machine type. This can be set in the build_opts file, e.g.
by adding the following line:
NUMBER_BITS_PER_SET = <number>

Change-Id: I314a3cadca8ce975fcf4a60d9022494751688e88
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18968
Reviewed-by: Tiago Mück <tiago.muck@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

base: Fix missing headers to CircularQueue

CircularQueue is currently throwing compilation errors when creating
a derived class.

assert() needs <cassert>
ptrdiff_t needs <cstddef>
(u)intX_t need <cstdint>
random_access_iterator_tag needs <iterator>
is_same, enable_if and others need <type_traits>

Change-Id: I77a78e7b13f7a8b8e7e8b2b872065d78d1ab815a
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19089
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

Revert "mem-cache: Remove writebacks packet list"

This reverts commit bf0a722acdd8247602e83720a5f81a0b69c76250.

Reason for revert: This patch introduces a bug:

The problem here is that the insertion of block A may cause the
eviction of block B, which on the lower level may cause the
eviction of block A. Since A is not marked as present yet, A is
"safely" removed from the snoop filter

However, by reverting it, using atomic and a Tags sub-class that
can generate multiple evictions at once becomes broken when using
Atomic mode and shall be fixed in a future patch.

Change-Id: I5b27e54b54ae5b50255588835c1a2ebf3015f002
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19088
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

cpu: Added the Multiperspective Perceptron Predictor (8KB and 64KB)

Described by the following article:
  Jiménez, D. "Multiperspective perceptron predictor."
  Championship Branch Prediction (CBP-5) (2016).

Change-Id: Iaa68ead7696e0b6ba05b4417d0322e8053e10d30
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/15495
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Ilias Vougioukas <ilias.vougioukas@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

x86: fix movsd bug on %xmm register

The movsd instruction should zero out half the register, but
does not do it. This changeset adds the necessary microop to
the instruction to cause correct behavior.

Change-Id: I5278da3634c78a97ed0586f687a36c6dc5a34c60
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19068
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Michael LeBeane <Michael.Lebeane@amd.com>
Reviewed-by: Gabe Black <gabeblack@google.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Gabe Black <gabeblack@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>

config, arm: memoryMode test

Commit 9a13acaa367769c38859342de9bc35aac59a6710 doesn't comply with the
same behaviour. The conditional test the memory mode on the same cluster,
while it should test the other cluster.

Change-Id: If3a7863f0905e66a548001d8e74689f5dd07179c
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/17748
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>

arm: Fix decoding of CRC32 instructions in thumb32

The CRC32 and CRC32C instructions are incorrectly decoded in thumb32
mode according to the latest manual:
https://developer.arm.com/docs/ddi0597/latest/top-level-encodings-for-t32/16-bit#dpint_2r

Change-Id: I9c6684f1ec7fe14d3b4cdf13f117a9819e046578
Signed-off-by: Chun-Chen TK Hsu
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/19028
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Giacomo Travaglini <giacomo.travaglini@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

cpu-o3: Increase LSQ buffer sizes to match max vector length

Change-Id: I5890c7cfa147125ce3389001f85d56d4b5a9911d
Signed-off-by: Gabor Dozsa <gabor.dozsa@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/13525
Tested-by: kokoro <noreply+kokoro@google.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Michael LeBeane <Michael.Lebeane@amd.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>

arch-arm: Treat SVE prefetch instructions as no-ops

Change-Id: Ife0424e274dd65d6dc4f6e5cc5e37d17b03be0d8
Signed-off-by: Giacomo Gabrielli <giacomo.gabrielli@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/13522
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>

arch-arm: Add initial support for SVE gather/scatter loads/stores

Change-Id: I891623015b47a39f61ed616f8896f32a7134c8e2
Signed-off-by: Giacomo Gabrielli <giacomo.gabrielli@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/13521
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

cpu: Fix rescheduling of progress check events

noRequestEvent needs to be rescheduled on recvRetry, otherwise the timeout
may be triggered even though packets are being eventually sent.
noResponseEvent scheduling is also fixed. This timeout should not be
active when we are not expecting a response.

Change-Id: If9edb75b5b803caf9f99bf41ea3948b15a3f3d71
Signed-off-by: Tiago Muck <tiago.muck@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18793
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch: Add include guards to auto-gen. decode header

Change-Id: I03bfc9035b82bc1a42e799cf645d43cb5dafb4cb
Signed-off-by: Giacomo Gabrielli <giacomo.gabrielli@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18911
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>

cpu-o3: Add support for pinned writes

This patch adds support for pinning registers for a certain number of
consecutive writes.  This is only relevant for timing CPU models
(functional-only models are unaffected), and it is primarily needed to
provide a realistic execution model for micro-coded operations whose
microops can write to non-overlapping portions of a destination
register, e.g. vector gather loads.  In those cases, this mechanism
can disable renaming for a sequence of consecutive writes, thus making
the resulting execution more efficient: allocating a new physical
register for each microop would introduce a read-modify-write chain of
dependencies, while with these modifications the microops can write
back in parallel.

Please note that this new feature is only leveraged by O3CPU for the
time being.

Additional authors:
- Gabor Dozsa <gabor.dozsa@arm.com>

Change-Id: I07eb5fdbd1fa0b748c9bdc1174d9f330fda34f81
Signed-off-by: Giacomo Gabrielli <giacomo.gabrielli@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/13520
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>

arch, base, cpu, gpu, sim: Merge getMemProxy and getVirtProxy.

These two functions were performing the same function but had two
different names for historical reasons. This change merges them
together, keeping the getVirtProxy name to be consistent with the
getPhysProxy method used to get a non-translating proxy port.

Change-Id: Idd83c6b899f9343795075b030ccbc723a79e52a4
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18581
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>

cpu: Store the translating proxy with the same pointer in SE or FS mode.

Only one is active at a time, so they can share the same pointer.

Change-Id: Ie4ae1f0ffbf9448f6730f9c7d072bc85d6d423da
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18580
Tested-by: kokoro <noreply+kokoro@google.com>
Maintainer: Gabe Black <gabeblack@google.com>
Reviewed-by: Brandon Potter <Brandon.Potter@amd.com>

cpu, sim: Return PortProxy &s from all the proxy accessors.

This is a step towards merging the accessors for SE and FS modes.

Change-Id: I76818ab88b97097ac363e243be9cc1911b283090
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18579
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Brandon Potter <Brandon.Potter@amd.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>

kern: Replace an explicitly instantiated port proxy with one from the tc.

That avoids having to know what type of proxy to create.

Change-Id: I311e770ab720061b52f29df0dcc2273e028aa34a
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18578
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Brandon Potter <Brandon.Potter@amd.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>

arch, base, sim: Demote (SE|FS)TranslatingPortProxy &s to PortProxy &s.

Al(most) all of the interesting differences between the two classes
have been removed. There are some control methods which are still
specific to each type which may require treating them as their true
type, but most code that consumes them doesn't need to worry about
which is which.

Change-Id: Ie592676f1e496c7940605b66e55cd7fae18e59d6
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18577
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Brandon Potter <Brandon.Potter@amd.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>

mem: Remove the now unused Copy* methods from the FS port proxy.

Change-Id: Ie433a9e4c9ee748911060eb7b1b47e617aa297a6
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18576
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Brandon Potter <Brandon.Potter@amd.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>

arch, base, sim: Replace Copy(String)?(In|Out) with equivalent code.

This expands those functions into code which extracts the virt proxy
and then uses the appropriate method on it. This has two benefits.
First, the Copy* functions where mostly redundant wrappers around the
methods the proxy port already had. Second, using them forced a
particular port which might not actually be what the user wanted.

Change-Id: I62084631dd080061e3c74997125164f40da2d77c
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18575
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

sim-se: const for loader's loadSection param

The port proxy can be declared as a reference to a const proxy
rather than just a reference to a proxy.

Change-Id: I4640b0c5f33e2334c1e7630131f78607ced40a34
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/12301
Maintainer: Brandon Potter <Brandon.Potter@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>

cpu: Added correct return type for ROB::countInsts

- return size_t (unsigned) according to the .size() return type
- fixed typo in doc (source of warning with some compilers)

Change-Id: I48ee2e317cf41011a6fcb5ca45aef67e75329bfa
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18948
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

mem-cache: Accuracy-based rate control for prefetchers

Added a mechanism to control the number of prefetches generated
based in the effectiveness of the prefetches generated so far.

Change-Id: I33af82546f74a5b5ab372c28574b76dd9a1bd46a
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18808
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>

sim-se: add a release parameter to Process.py

Set the default release to that single value for all ISAs.

glibc has checks for the kernel version based on uname, and refuses
to start any syscall emulation programs if those checks don't pass with
error:

FATAL: kernel too old

The ideal solution to this problem is to actually implement all missing
system calls for the required kernel version and bumping the release
accordingly.

However, it is very hard to implement all missing syscalls and verify
compliance.

Previously, we have simply bumped the version manually from time to
time when major glibc versions started breaking.

This commit alleviates the problem in two ways.

Firstly, having a single kernel version for all versions means that it is
easier to bump all versions at once.

Secondly, it makes it is possible to set the release with a parameter,
which in turn can be set from the command line with:

se.py --param 'system.cpu[:].workload[:].release = "4.18.0"'

Change-Id: I9e3c31073bfe68735f7b0775c8e299aa62b98222
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/17849
Maintainer: Brandon Potter <Brandon.Potter@amd.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>

mem-cache: Support for page crossing prefetches

Prefetchers can now issue hardware prefetch requests that go beyond
the boundaries of the system page. Page crossing references will need
to look up the TLBs to be able to compute the physical address to be
prefetched.

Change-Id: Ib56374097e3b7dc87414139d210ea9272f96b06b
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/14620
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>

mem: Add a readString method to the PortProxy which takes a char *.

This version takes a char * instead of an std::string &, and a maximum
length to fill in like strncpy. This is intended to be a replacement
for the CopyStringOut function.

Change-Id: Ib661924a3fa7e05761d572ffecbe2c0cc8659d48
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18574
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Brandon Potter <Brandon.Potter@amd.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>

mem: Use a const T & in write<> to avoid an unnecessary copy.

If the type T is complex/large, the it makes sense to access it in place
and not copy it and then not modify it.

Change-Id: Idd24be4fbba636375637ff72b1ba5ee32eb76215
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18573
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>

arch, base, dev, sim: Remove now unnecessary casts from PortProxy methods.

Change-Id: Ia73b2d86a10d02fa09c924a4571477bb5f200eb7
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18572
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>

mem, arm: Replace the pointer type in PortProxy with void *.

The void * type is for pointers which point to an unknown type. We
should use that when handling anonymous buffers in the PortProxy
functions, instead of uint8_t * which points to bytes.

Importantly, C/C++ doesn't require you to do any casting to turn an
arbitrary pointer type into a void *. This will get rid of lots of
tedious, verbose casting throughout the code base.

Change-Id: Id1adecc283c866d8e24524efd64f37b079088bd9
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18571
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>

mem, arm: Move some helper methods into the base PortProxy class.

These were originally in the SETranslatingPortProxy class, but they're
not specific to SE mode in any way and are an unnecessary divergence
between the SE and FS mode translating port proxies.

Change-Id: I8cb77531cc287bd15b2386410ffa7b43cdfa67d0
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18570
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>

arm, mem: Move the SecurePortProxy subclass into it's own file.

The idea of a "secure" memory area/access is specific to ARM and
shouldn't be in the common mem directory, although it's built in to the
generic memory protocol at this point.

Regardless, it should minimially be in its own file like the virtual
and physical port proxy classes are.

Change-Id: I140d4566ee2deded784adb04bcf6f11755a85c0c
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18569
Reviewed-by: Gabe Black <gabeblack@google.com>
Maintainer: Gabe Black <gabeblack@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>

mem: Parameterize coherent xbar sanity checks

Parameters can be used to change coherent xbar limits for the routing
table and outstanding snoops. We need the ability to tweak these values
as the current defaults may be violated in simulations with large core
counts.

Change-Id: Idb64b8c105683d02d8beba5bce13b815181ba824
Signed-off-by: Tiago Muck <tiago.muck@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/18789
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>