By default the GPU VIPER coherence protocol uses a WT L2 cache.
However it has support for using WB caches (although this is not
tested currently). When using a WB L2 cache for the GPU, this
results in deadlocks with atomics.
Specifically, when an atomic reaches the L2 and the line is
currently in M or W, the line must be written back before the atomic
can be performed. However, the current support has two issues:
a) it never performs the atomic operation -- while VIPER current
assumes all atomics are system scope atomics and thus cannot be
performed at the L2 and this transition requires the dirty line be
written back before performing the atomic, the transition never
performs the atomic nor does the response path handle it.
b) putting the atomic action right after the write back is not
safe because we need to ensure the requests are ordered when they
reach memory -- thus we have to wait until the write back is
acknowledged before it's safe to send/perform the atomic.
To fix this, this change modifies the transition in question to
put the atomic on the stalled requests buffer, which the WBAck will
check when it returns to the L2 (and thus perform the atomic, which
will result in the atomic being sent on to the directory).
This fix has been tested and verified with both the per-checkin and
nightly GPU Ruby Random tester tests (with a WB L2 cache).
Change-Id: I9a43fd985dc71297521f4b05c47288d92c314ac7
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/68978
Maintainer: Bobby Bruce <bbruce@ucdavis.edu>
Reviewed-by: Matthew Poremba <matthew.poremba@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
By default the GPU VIPER coherence protocol uses a WT L2 cache.
However it has support for using WB caches (although this is not
tested currently). When using a WB L2 cache for the GPU, this
results in deadlocks with loads.
Specifically, when a load reaches the L2 and the line is currently
in the W state, that line must be written back before the load can
be performed. However, the current transition for this in the L2
did not attempt to retry the load when the WB completes, resulting
in a deadlock. This deadlock can be replicated by running the GPU
Ruby random tester as is with a WB L2 cache instead of a WT L2
cache.
To fix this, this change modifies the transition in question to
put the load on the stalled requests buffer, which the WBAck will
check when it returns to the L2 (and thus perform the load).
This fix has been tested and verified with both the per-checkin and
nightly GPU Ruby Random tester tests (with a WB L2 cache).
Change-Id: Ieec4f61a3070cf9976b8c3ef0cdbd0cc5a1443c6
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/68977
Reviewed-by: Matthew Poremba <matthew.poremba@amd.com>
Maintainer: Bobby Bruce <bbruce@ucdavis.edu>
Tested-by: kokoro <noreply+kokoro@google.com>
The DEBUG macro is not part of any compiler standards (differently from
NDEBUG, which elides assertions).
It is only meant to differentiate gem5.debug from .fast and .opt builds.
gem5 developers have used it to insert helper code that is supposed to
aid the debugging process in case anything goes wrong.
This generic name is likely to clash with other libraries linked with
gem5. This is the case of DRAMSim as an example.
Rather than using undef tricks, we just inject a GEM5_DEBUG macro
for gem5.debug builds.
Change-Id: Ie913ca30da615bd0075277a260bbdbc397b7ec87
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/69079
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
This piggy-backs on the writeOK method which already exists. It also
modifies the flags returned as part of the memory's backdoor
descriptor which doesn't enforce that the memory is read only, but will
let the other party know it's expected not to write to it.
Change-Id: Ib95e619c76c327d302e62a88515a92af11815981
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/68557
Maintainer: Gabe Black <gabeblack@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matthew Poremba <matthew.poremba@amd.com>
The reference can be bound to an invalid object (*nullptr) in
situations where no proper owner SimObject can be provided to the port
constructor. This rightfully triggers a UBSAN warning.
Also, these two classes do not make use of the owner reference member
themselves and expose it as a protected member reference to
subclasses. This desing has several drawbacks: requires the reference
to owner to travel the class hierarchy up and down, loosing its true
static type in the process ; non-private member variable should not be
part of the API of such fundamental classes, if only for
maintainability ; a reference bound from a nullable pointer is a lying
API as it hides the optional aspect of ownership.
Note that the reference to invalid object can't be properly fixed until
the complete removal of the owner reference. This patch lays the path
toward that fix.
Change-Id: I8b42bc57d7826656726f7708492c43366f20633a
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/67551
Reviewed-by: Bobby Bruce <bbruce@ucdavis.edu>
Tested-by: kokoro <noreply+kokoro@google.com>
Maintainer: Bobby Bruce <bbruce@ucdavis.edu>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
We now explicitly check in both the cache and the MSHRs if writes are
masked or not before promoting to a whole-line write. Failure to do
this previously was resulting in data loss when dirty data was present
in lower level caches and a coincidentally aligned and
cache-line-sized masked write occured.
Change-Id: I9434590d8b22e4d993167d789eb9d15a2e866bf1
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/64340
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
The port_wrapper classes convert the Request/ResponsePort from
inherit-base to callback registrations. This help 'composition over
inheritance' that most design pattern follows, which help reducing
code length and increase reusability.
Change-Id: Ia13cc62507ac8425bd7cf143a2e080d041c173f9
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/67232
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
66d4a158 added support for AMD's GPU cache bypassing flags (GLC
for bypassing L1 caches, SLC for bypassing all caches). However,
it did not add a transition for the situation where the cache line
is currently I (Invalid). This commit adds this support, which
resolves an assert failure in Pannotia workloads when this situation
arises.
Change-Id: I59a62ce70c01dd8b73aacb733fb3d1d0dab2624b
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/67201
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
66d4a158 added support for AMD's GPU cache bypassing flags (GLC
for bypassing L1 caches, SLC for bypassing all caches). However,
for applications that use the GLC flag but intermix GLC- and
non-GLC accesses to the same address, this previous commit
has a bug. This bug manifests when the address is currently
valid in the L1 (TCP). In this case, the previous commit chose
to evict the line before letting the bypassing access to proceed.
However, to do this the previous commit was using the inv_invDone
action as part of the process of evicting it. This action is only
intended to be called when load acquires are being performed
(i.e., when the entire L1 cache is being flash invalidated). Thus,
calling inv_invDone for a GLC (or SLC) bypassing request caused an
assert failure since the bypassing request was not performing a
load acquire.
This commit resolves this by changing the support in this case to
simply invalidate the entry in the cache.
Change-Id: Ibaa4976f8714ac93650020af1c0ce2b6732c95a2
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/67199
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Two W->WI transitions, on events RdBlk and Atomic in the GPU L2 cache
coherence protocol do not clear the request from the request queue upon
completing the transition. This action is not performed in the respone
path. This update adds the p_popRequestQueue action to each of these
transitions to remove the stale request from the queue.
Change-Id: Ia2679fe3dd702f4df2bc114f4607ba40c18d6ff1
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/67192
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
An earlier commit added support for GLC and SLC AMDGPU instruction
modifiers. These modifiers enable cache bypassing when set. The GLC/SLC
flag information was being threaded through all the way to memory and
back so that appropriate actions could be taken upon receiving a request
and corresponding response. This commit removes the threading and adds
the bypass flag information to TBE. Requests populate this
entry and responses access it to determine the correct set of actions to
execute.
Change-Id: I20ffa6682d109270adb921de078cfd47fb4e137c
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/67191
Maintainer: Matt Sinclair <mattdsinclair@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
The GPU cache models do not support cache bypassing when the GLC or SLC
AMDGPU instruction modifiers are used in a load or store. This commit
adds cache bypass support by introducing new transitions in the
coherence protocol used by the GPU memory system. Now, instructions with
the GLC bit set will not cache in the L1 and instructions with SLC bit
set will not cache in L1 or L2.
Change-Id: Id29a47b0fa7e16a21a7718949db802f85e9897c3
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/66991
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Matt Sinclair <mattdsinclair@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
An example case,
```python
mem_side_port = RequestPort(
"This port sends requests and " "receives responses"
)
```
This is the residue of running the python formatter.
This is done by finding all tokens matching the regex `"\s"(?![.;"])`
and manually replacing them by empty strings.
Change-Id: Icf223bbe889e5fa5749a81ef77aa6e721f38b549
Signed-off-by: Hoa Nguyen <hoanguyen@ucdavis.edu>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/66111
Reviewed-by: Bobby Bruce <bbruce@ucdavis.edu>
Maintainer: Bobby Bruce <bbruce@ucdavis.edu>
Tested-by: kokoro <noreply+kokoro@google.com>
Make this part of the Functional protocol, since it should always
return immediately, can be shared by the atomic and timing protocols,
and thematically fits with that protocol.
The default implementation on the receiving end just ignores the
request and leaves the back door pointer set to null, effectively
making back doors default "off" which matches their behavior in the
atomic protocol.
This mechamism helps fix a bug in the TLM gem5 bridges which need to
translate to/from the DMI and back door mechanisms, where there can be
an explicit request for a back door which does not have a transaction
associated with it. It is also necessary for bridging DMI requests in
timing mode, since the DMI requests must be instant, and the timing
protocol does not send/receive packets instantly.
Change-Id: I905f13b9bc83c3fa7877b05ce932e17c308125e2
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/65752
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Gabe Black <gabeblack@google.com>
In cases where a back door is not being requested alongside a packet
or request, there needs to be a structure which describes the address
range to use, and what type of access the back door should support. It
would be possible to make a Packet/Request to carry that information,
but those types are actually pretty big, and have a lot of extra
overhead which would be overkill for this purpose.
Change-Id: I3638361ffa758ee959cb3bc57f7c35f2aa34a36c
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/65751
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Gabe Black <gabeblack@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Previously, shared memory server remove old socket *before* filling the
target path into API's data structure. However, the target path might
get truncated hence the path we check against might not be the one we
will be using in the end.
In a case where the path specified by user is free while the truncated
path is in used, gem5 will get a mysterious EADDRINUSE.
We swap the two steps in the CL, so we'll be checking against the actual
path we use, instead of the path user request to use.
Change-Id: Ib34f8b00ea1d2f15dcd4e7b6d2d4a6d6ddc4e411
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/65153
Reviewed-by: Gabe Black <gabeblack@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Maintainer: Gabe Black <gabeblack@google.com>
The current MESI_Two_Level protocol's L1 caches updates the MRU information twice per request on misses -- once when the request reaches Ruby and once when the miss is returned from another level of the memory hierarchy.
Although this approach does not cause any correctness bugs for replacement policies like LRU since this request is the LRU in both cases, it does not work correctly for other policies like SecondChance and LFU, where updating the information twice (for misses) causes them to devolve to LRU.
Note that this was not directly a problem with Ruby previously, because it only supported LRU-based policies that were unaffected by this. However, with the integration of 20879 Ruby now uses the same replacement policies as Classic (which has additional, non-LRU based replacement policies).
This patch resolves this problem by not updating the MRU information a second time for the misses. It has been tested and validated with the replacement policy tests.
Change-Id: I9e7e96a9d6c09f3d6b7daae7115ef091ac3bdc08
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/64371
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Maintainer: Matt Sinclair <mattdsinclair@gmail.com>
Clang Version 14 throws a warning "use of bitwise '&/|' with boolean
operands" for cases where bitwise operations are used where boolean
operations are intended.
This occurred in "WriteMast.hh", "data.isa", and "decode.cc" where
boolean values were being compared using the bitwise operands. While
bitwise operations are equivalent, they have been changed to boolean
operations in this patch to avoid the clang-14 warning.
Change-Id: Ic7583e13a325661712c75c8e1b234c4878832352
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/64172
Reviewed-by: Tom Rollet <tom.rollet@huawei.com>
Reviewed-by: Kunal Pai <kunpai@ucdavis.edu>
Maintainer: Bobby Bruce <bbruce@ucdavis.edu>
Tested-by: kokoro <noreply+kokoro@google.com>
The `Addr line_addr` in "src/mem/snoop_filter.cc" variable was only
used in an assert, stripped when compiling gem5.fast.
Clang-13 throws a warning for this variable. This has been fixed by
merging the variable and associated logic into the assert statement.
The variables in inet.cc and Sequencer.cc were also causing an 'unused
variable' warning to be thrown due to variables that were only used in
assert statements. In these cases the logic could not be moved into the
assert statement and, as such, the `GEM5_VAR_USED` MACRO is used to
remove this warning.
Change-Id: I6511d0863608c38b79e4558c7dcf35a323fe8362
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/64171
Reviewed-by: Kunal Pai <kunpai@ucdavis.edu>
Tested-by: kokoro <noreply+kokoro@google.com>
Maintainer: Bobby Bruce <bbruce@ucdavis.edu>
AddrRangeMap::intersects doesn't support ranges with different
interleavings, thus the current implementation of the destination
seach won't work in cases when different machines map the same address
with different interleaving.
The fixed implementation uses a different AddrRangeMap for each mach
type.
Change-Id: Idd0184da343c46c92a4c86f142938902096c2b1f
Signed-off-by: Tiago Mück <tiago.muck@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/63671
Tested-by: kokoro <noreply+kokoro@google.com>
Maintainer: Bobby Bruce <bbruce@ucdavis.edu>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Finish_CopyBack_Stale is scheduled only when the requestor is the last
sharer. This prevents the cacahe evicting the line which was already
evicted while the stale WriteBack transaction was stalled.
Wrong condition check in Finish_CopyBack_Stale for eviction is also
removed.
Change-Id: Ib66acc1b9e4a6f7cea373e1fb37375427897d48d
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/63611
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
