Enable the cache to detect contiguous writes and hold on to the MSHR
long enough to allow the entire line to be written. If the whole line
is written, the MSHR will be sent out as an invalidation requests, as
it is part of a whole-line write, i.e. no-fetch-on-write.
The cache is also able to switch to a write-no-allocate policy on the
actual completion of the writes, and instead use the tempBlock and
turn the write operation into a writeback.
These policies are all well-known, and described in works such as
Jouppi, Cache Write Policies and Performance, vol 21, no 2, ACM, 1993.
Change-Id: I19792f2970b3c6798c9b2b493acdd156897284ae
Reviewed-on: https://gem5-review.googlesource.com/c/12907
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
An MSHR is allocated and the computed latency determines when the MSHR
will be ready and can be serviced by the cache. This patch adds a
function that allows changing the time that an MSHR is ready and
adjusts the queue such that other MSHRs can be serviced first if they
are ready.
Change-Id: Ie908191fcb3c2d84d4c6f855c8b1e41ca5881bff
Reviewed-on: https://gem5-review.googlesource.com/c/12906
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
This patch changes how we deal with whole-line writes their
responses. With these changes, we use the MSHR tracking to determine
if a whole-line is written, and on a fill we simply handle the
invalidation response, with the actual writes taking place as part of
satisfying the CPU-side hit.
Change-Id: I9a18e41a95db3c20b97f8bca7d95ff33d35a578b
Reviewed-on: https://gem5-review.googlesource.com/c/12905
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
This patch adds support for determining whether the targets in an MSHR
are 1) only writes and 2) whether these writes are effectively a
whole-line write. This patch adds the necessary functions in the MSHR
to allow for write coalescing in the cache.
Change-Id: I2c9a9a83d2d9b506a491ba5b0b9ac1054bdb31b4
Reviewed-on: https://gem5-review.googlesource.com/c/12904
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
The NULL ISA doesn't really have an endianness. Now that the packet
accessors which consumed that endianness are gone, we can get rid of
that setting as well.
Change-Id: I8dd4c7b8236b07df4458fea377865f30141121d4
Reviewed-on: https://gem5-review.googlesource.com/c/13466
Reviewed-by: Gabe Black <gabeblack@google.com>
Maintainer: Gabe Black <gabeblack@google.com>
The accessors are used for debugging output. If we're using an ISA
where there's an endianness, we use that explicitly, falling back to a
binary dump if the size isn't supported. If not, then we just dump the
data without interpretation regardless of size.
Change-Id: Ib050c4c876ee41f17cfd14ad657150bf6ab1de39
Reviewed-on: https://gem5-review.googlesource.com/c/13464
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Gabe Black <gabeblack@google.com>
This avoids a place where data has its endianness switched so that when
the endianness based accessors switch it back it returns to normal. It
also makes it easier to show intent when accessing single bytes where
endianness doesn't matter, and there's no contextual endianness.
Change-Id: I1b97396c1b9bb39727d35112d90e3969e5fe0aab
Reviewed-on: https://gem5-review.googlesource.com/c/13455
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
ReplaceableEntry is referenced by many classes that do
not necessarily need access to the replacement policies.
Therefore, in order to allow better compilation units,
we factor it out to a new file.
Change-Id: I0823567bf1ca336ffcdf783682ef473e8878d7fd
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/13418
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Implementation of a Tree-PLRU replacement policy. It is based on
the assumption that a set associative cache is used.
Change-Id: I74b227e88fd6c93aab5bb2cd0e8730376db28f52
Reviewed-on: https://gem5-review.googlesource.com/c/11106
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Replacement policies aren't aware of cache sets and do not
organize blocks based on replacement data. Block search is
independent of block placement.
Besides, indexing policies have their own way of addressing
the sets, therefore there is no need to use this class anymore.
BlkType has been removed, as it wasn't being used.
Change-Id: Ia79c2a491e59f295c8d60a0466c317eb0e2bdab9
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/9782
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Split indexing functionality from tags, so that code duplication
is reduced when adding new classes that use different indexing
policies, such as set associative, skewed associative or other
hash-based policies.
An indexing policy defines the mapping between an address' set
and its physical location. For example, a conventional set assoc
cache maps an address to all ways in a set using an immutable
function, that is, a set x is always mapped to set x. However,
skewed assoc caches map an address to a different set for each way,
using a skewing function.
FALRU has been left unmodified as it is a specialization with its
own complexity.
Change-Id: I0838b41663f21eba0aeab7aeb7839e3703ca3324
Reviewed-on: https://gem5-review.googlesource.com/c/8885
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Replaceable entries belong to table-like structures, and therefore
they should be indexable by combining a row and a column. These,
using conventional cache nomenclature translate to sets and ways.
Make these entries aware of their sets and ways. The idea is to
make indexing policies usable by other table-like structures. In
order to do so we move sets and ways to ReplaceableEntry, which
will be the common base among table entries.
Change-Id: If0e3dacf9ea2f523af9cface067469ccecf82648
Reviewed-on: https://gem5-review.googlesource.com/c/12764
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Having the blocks initialized in the constructor makes it harder
to apply inheritance in the tags classes. This patch decouples
the block initialization functionality from the constructor by
using an init() function. It also sets the parent cache.
Change-Id: I0da7fdaae492b1177c7cc3bda8639f79921fbbeb
Reviewed-on: https://gem5-review.googlesource.com/c/11509
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Decouple Tags from Packets, only extracting the necessary
functionality for block insertion. As a side effect, create
a new function to update common insertion statistics.
Change-Id: I5c58f7c17de3255beee531f72a3fd25a30d74c90
Reviewed-on: https://gem5-review.googlesource.com/c/11098
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
The block was being invalidated before the hash could erase
its entry, therefore it was using invalid values (tag was
being assigned MaxAddr and the secure bit was reset).
This change reorders the calls, so that the appropriate hash
entry is erased.
Change-Id: I161463df0f8f5220179bc68d7be12051e5390d01
Signed-off-by: Daniel R. Carvalho <odanrc@yahoo.com.br>
Reviewed-on: https://gem5-review.googlesource.com/c/13210
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
In the previous implementation, messages are randomly inserted with
delays only if both RubySystem and MessageBuffer randomization flags
are set true. However, to find race conditions and cover more slicc
transitions, ruby random testers rely on setting RubySystem flag to
turn on randomization on all message buffers.
As a fix, this patch enables a message buffer to have randomization
when either RubySystem or its own flag is set.
Change-Id: I1e076908ff07e5846ebad4f4fc1c8f28d40bbfd4
Reviewed-on: https://gem5-review.googlesource.com/12784
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
The hash() function must be bijective for the skewed caches to work,
however when the hashing is done on top of a one-bit address, the
MSB and LSB refer to the same bit, and therefore their xor will
always be zero.
This patch adds a fatal error to not allow the user to set an invalid
value for the number of sets that would generate that bug.
As a side note, the missing header for the bitfields functions has
been added.
Change-Id: I35a03ac5fdc4debb091f7f2db5db33568d0b0021
Reviewed-on: https://gem5-review.googlesource.com/12724
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Providing a configurable fair scheduling policy based on utilization;
utilization is directly proportional to a score which is inversely
proportional to the QoS priority. It is meant to avoid starvation of low
priority packets.
Users can tune the policy by adjusting the weight parameter (weight of
the following formula)
new_score = ((1.0 - weight) * old_score) + (weight * served_bytes)
Change-Id: I7679e234b916c57ebed06cec0ff3cff3cf2aef22
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/12359
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
"4976ff5 mem-cache: Refactor the recvAtomic function" introduced a bug
where if an atomic request that fills in using the tempBlock it will
not evict it when it finishes handling the request as it should. This
triggers an assertion. This change fixes this bug.
Change-Id: I73c808a7e15237eddb36b5448ef6728f7bcf7fd9
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/12644
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
This is the implementation of QoS algorithms support for gem5 memory
objects. This change-list provides a framework for specifying QoS
algorithm which can be used to prioritise service to specific masters in
the memory controller.
The QoS support implemented here is designed to be extendable so that
new QoS algorithms can be easily plugged into the memory controller as
"QoS Policies".
Change-Id: I0b611f13fce54dd1dd444eb806f8e98afd248bd5
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/11970
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Create a class that implements the skewed associative placement
policy. It uses the hash function and expansions of the skewing
functions described in "Skewed-Associative caches", by Seznec.
Only 8 skewing functions are implemented, and therefore if more
are needed a hash function will be recursively applied on top
of the output of one of these functions to generate different
values. This is not optimal, and if more functions are needed
it might be more effective to implement them.
Change-Id: Ibc77edffd8128114a8b200cec5d8deedfb5105cb
Reviewed-on: https://gem5-review.googlesource.com/8886
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
This patch adds StreamID and SubstreamID to Request. These fields can
be used by a SMMU/IOMMU model to pick up the correct translation
context for each request and they correspond to an ASID in a device.
For this reason they have been merged together with the request asid
in a union, so that a cpu will set the asid and a device will set
the Stream and Substream ID.
Change-Id: Iac2b5a1ba9c6598ee7635c30845dc68ba6787c34
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/12187
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Do not generate garnet tester file or Ruby debug headers without a Ruby
protocol (i.e. PROTOCOL=None). It makes no sense to include these files
into the build when there will be no protocol to utilize them.
Change-Id: I8db4dd532f60008217a10c88a2e089f85df9d104
Reviewed-on: https://gem5-review.googlesource.com/8381
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
The commented code contains bitrot. It is not clear how to fix the code
so remove it.
The code will not compile if the preprocessor defines are removed. The
llocker and uulocker variables that are used as indices into the
persistent_randomize array are undefined. It's not clear what they should
be from the current code.
5ab13e2deb shows when the lines were last modified. The functionality
contained in the comments probably have not been used since that time.
(This is an example of why one should never add commented code that
is enabled by removing defines. The code rots and sits in the source
forever.)
Change-Id: I3e0e7c9afc0b6088130e6f319075809fb6f16e5a
Reviewed-on: https://gem5-review.googlesource.com/8481
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
This change is because I want to make CacheBlk::data private, so that
I can track all the places which write to it. But to keep that commit
smaller (it is pretty big, because of all the places which might
change it), I have split this into a commit of its own.
Change-Id: I15a2fc1752085ff3681f5c74ec90be3828a559ea
Reviewed-on: https://gem5-review.googlesource.com/11829
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Packet::checkFunctional also wrote data to/from the packet depending
on if it was read/write, respectively, which the 'check' in the name
would suggest otherwise. This renames it to doFunctional, which is
more suggestive. It also renames any function called checkFunctional
which calls Packet::checkFunctional. These are
- Bridge::BridgeMasterPort::checkFunctional
- calls Packet::checkFunctional
- MSHR::checkFunctional
- calls Packet::checkFunctional
- MSHR::TargetList::checkFunctional
- calls Packet::checkFunctional
- Queue<>::checkFunctional
(of src/mem/cache/queue.hh, not src/cpu/minor/buffers.h)
- Instantiated with Queue<WriteQueueEntry> and Queue<MSHR>
- WriteQueueEntry
- calls Packet::checkFunctional
- WriteQueueEntry::TargetList
- calls Packet::checkFunctional
- MemDelay::checkFunctional
- calls QueuedSlavePort/QueuedMasterPort::checkFunctional
- Packet::checkFunctional
- PacketQueue::checkFunctional
- calls Packet::checkFunctional
- QueuedSlavePort::checkFunctional
- calls PacketQueue::doFunctional
- QueuedMasterPort::checkFunctional
- calls PacketQueue::doFunctional
- SerialLink::SerialLinkMasterPort::checkFunctional
- calls Packet::doFunctional
Change-Id: Ieca2579c020c329040da053ba8e25820801b62c5
Reviewed-on: https://gem5-review.googlesource.com/11810
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
To avoid unintentional variable capture, all std calls must be
prefixed. These are the identifiers which are in the std
namespace (according to
https://en.cppreference.com/w/cpp/symbol_index), but that will remain
unprefixed with this change:
int8_t int16_t int32_t int64_t
uint8_t uint16_t uint32_t uint64_t
The (u)int types are included from the packet header file, which
includes <inttypes.h>, where they occur in the global namespace. They
are in the std namespace in <cinttypes>/<cstdint>.
There is an occurrence of "set" in this file, which is "Packet::set"
and not "std::set", so it is not prefixed with the std namespace
Change-Id: I7f6c0b61b09658e224fe31a9f73150b81861d6f8
Reviewed-on: https://gem5-review.googlesource.com/11809
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
There was an off-by-one error in the isRead() case, as `val_end` and
`func_end` pointed to the last byte to write to (not one past the last
byte), and thus `*_end - *_start` was not the length of the data to
memcpy.
This was correct in the case of
val_start >= func_start && val_end <= func_end
where `overlap_size = size`, but if it were (as the other cases
suggest) `overlap_size = val_end - val_start`, then it would also be
off by one.
Also, the isWrite() case catered for this.
I simplified the four ifs into one case which uses min/max (this is
how I spotted the inconsistency).
Change-Id: Ib5c5da084652e752f6baf1eec56b51b4f0f5c95c
Reviewed-on: https://gem5-review.googlesource.com/11750
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Add a memory system component that delays traffic. The base
functionality to delay packets is implemented in the abstract MemDelay
class. This class exposes three methods that control packet delays:
* delayReq(pkt)
* delayResp(pkt)
* delaySnoopResp(pkt)
These methods should be specialized to implement delays for specific
packet types.
The class SimpleMemDelay uses the MemDelay base class to implement
constant delays for read/write requests and responses.
The intention is that these classes can be used for rapid prototyping
of components that add a small fixed delay and the same throughput as
the interconnect. I.e., any buffering done in the base class will be
small and proportional to the introduced delay.
Change-Id: I158cb85f20e32bfdbcbfed66a785b4b2dd47b628
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Nicholas Lindsey <nicholas.lindsay@arm.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/11521
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
While a cache clean operation is pending, all requests to the
corresponding block get deferred. When the response of a cache clean
operation is received, if the block is present and the response is not
invalidating, we can service all deferred targets that didn't require
writable. This change implements this functionality.
Change-Id: Ief47e74d07749a6a9736ab450eb46eefa53464a2
Reviewed-on: https://gem5-review.googlesource.com/11018
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Until now, all deferred targets of an MSHR would be promoted together
as soon as the targets were serviced. Due to the way we handle cache
clean operations we might need to promote only deferred targets that
don't require writable, leaving some targets as deferred. This change
adds support for this selective promotion.
Change-Id: I502e523dc9adbaf394955cbacea8286ab6a9b6bc
Reviewed-on: https://gem5-review.googlesource.com/11017
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
There are cases where a request which does not need a writable copy
gets an response upgraded reponse and fills in a writable copy. When
this happens, we promote deferred MSHR targets that were deferred
because they needed a writable copy to service them immediately.
Previously, we would uncoditionally promote deferred targets. Since
the deferred targets might contain a cache invalidation operation, we
have to make sure that any targets following the cache invalidation is
not promoted.
Change-Id: I1f7b28f7d35f84329e065c8f63117db21852365a
Reviewed-on: https://gem5-review.googlesource.com/11016
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Until now, all deferred targets of an MSHR would be promoted together
as soon as the targets were serviced. When we promote deferred targets
we also clear the downstreamPending flag.
Due to the way we handle cache clean operations we might need to
promote only deferred targets that don't require writable, leaving
some targets as deferred. To allow for partial target promotion, this
change adds support for clearing the downstreamPending only for a
subset of a TargetsList.
Change-Id: Id06953643ba9a975ebacc76ac10215441e264e74
Reviewed-on: https://gem5-review.googlesource.com/11015
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
TempCacheBlock insert() had a different signature than the parent class
which caused an error on clang. This matches the signature with default
zero values.
Change-Id: Ic096914497f3d17e88295c9e65a04d76fdddf365
Signed-off-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-on: https://gem5-review.googlesource.com/11349
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Previously, we used the start address to determine the right physical
memory while servicing memory requests. This change uses the full
address range to correctly determine the right physical memory and
expose bugs where requests might not fully map to a single physical
memory.
Change-Id: I183d7552918106000f917a62ceb877511ff0ff71
Reviewed-on: https://gem5-review.googlesource.com/11118
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
