The transition that happens when TCC acknowledges TCP of an atomic
operation completion does not move the cacheline state from A to I. This
commit fixes the transition and moves the state to I
When the cache is performing an atomics and receives data, it performs.
pa_performAtomic. This action peeks into the coreRequest queue to check
the messaage type. This queue, however, is already dequeued in the
transition that precedes the one that contains pa_performAtomic. When
pa_performAtomic is called, the simulation crashes. This commit fixes
the crash by using the TBE entry information instead of peeking when TBE
entry exists, and peeking when it doesn't
Previously, a warning would be printed when the WPRI bits
in the senvcfg register were written to. The other registers
do not print warnings for this, so the warning is being
removed.
This commit adds the senvcfg CSR, which fixes the 6.11.3 kernel
crash documented in issue 1674. I have not added a bitfield and
its implementation in isa.cc only uses setMiscRegNoEffect, so
this implementation is likely missing some critical components.
The previous #484 issue reported a bug where the TLB stats on RISC-V
were incremented twice on misses by calling the `lookup` function twice
with hidden argument set to `false`. The fix is only applied on atomic
mode as the `translation` argument of `doTranslate` will not be
`nullptr` in timing mode.
In that case, if the TLB lookup miss, the `doTranslate` function will
start the walker and then return without doing anything more. Then
later, when the pagetable walker found the corresponding PTE, it will
insert it and call `translateWithTLB`. This function then call `lookup`
again which will hit in any case (and crash if not due to the following
assert), but the hit count is incremented here too.
This commit fix by setting the `hidden` argument of `lookup` to true.
GCC and CLANG have different annotations for declaring code should not
be optimized. Adding GEM5_NO_OPTIMZE provides gem5 developers a MACRO
that works in both cases.
This change replaces the GCC pragmas in vfp.hh with GEM5_NO_OPTIMIZE
as this solution didn't work with clang.
In MI_example, when in MI state the block "Maybe_Stale" as in this
controller may have the most up to date value or it could be in the
network. For MII it is guaranteed that this controller has the most up
to date value because it received a PUTX_NACK.
This fixes one of the daily test failures.
- In the new MultiRuby system, the generated ProtocolInfo header files were not being correctly added to the build targets in SCons.
- As a result, when building gem5 with the --duplicate-sources option, these files were mistakenly deleted by SCons.
This happened because SCons treated them as source files instead of generated build targets.
- This commit ensures that the ProtocolInfo header files are explicitly included in the build targets, preventing their unintended removal and fixing the build issue.
The test `ruby_test_test-ALL-x86_64-opt-MatchStdout` is currently
failing because the reference file doesn't match the actual output. This
PR changes the reference file to match.
Fix#1809. Shift the mmap end to a lower address in case the process has
a large max stack size, to avoid overlapping the stack with the mmap
memory range.
Change-Id: Idae343dbbe851a7510463ff141c03f1847e36328
Previous PR #1758 implements the generic getValidAddr to get pure
vaddr without any tags or sign-extend bits.
In RISC-V implementation, the getValidAddr will zero-extend
address in RV32 mode and use it to do TLB translation. Use
getValidAddr to get zero-extend vaddr can reduce zero-extend
repetition
Change-Id: I2273ce48bccb873790103ba0fcdb0b48de9ced4c
src/cpu/simple/probes/LooppointAnalysis.py:
- remove default values for bb_valid_addr_range and
marker_valid_addr_range
- add more comments to explain parameter behaviors
- add citation to the LoopPoint paper
src/cpu/simple/probes/looppoint_analysis.cc:
- fix the incorrect styles
- remove updateBackwardBranch() function call
- match the style of checking if listeners vector is empty
- change the way of stopListening() to remove the listeners through the
manager instead of through the ProbeListener object's destructor.
src/cpu/simple/probes/looppoint_analysis.hh:
- removed backwardBranchPC and use the backwardBranchCounter to replace
its functionaility. Therefore, also removed updateBackwardBranch
function.
Change-Id: Id2430e2f04e61f72d5c4f1aad5cfd4d24a0fbc45
Add comments to most variables and functions.
Change the naming of some variables and functions to improve the
clearness.
Change-Id: Idb557ec84698b4344ed4683f5de87b1a3c2fd66d
In looppoint_analysis.hh, added LooppointAnalysis and
LooppointAnalysisManager classes.
Added all functions and variables for the classes.
Comments needed.
Change-Id: Ia7425b672ef092a68c99b702136850bfa1fcf0a2
Because the LoopPoint analysis will be done with ATOMIC CPU, so all
files related to the LoopPoint analysis object will be under
/src/cpu/simple/probes.
Change-Id: Icbdb0742b712a23dc8f6a19f4c1c827a1f5bf288
This change updates the names for the GPU children in a better way than
overriding the parent. Now it looks something like
```text
board.gpus.shader.CUs00
board.gpus.gpu_caches.ruby_gpu.controllers02
board.gpus.memory.mem_ctrl0
```
Note that it is "gpus" with an "s" because the board accepts more than 1
GPU, optionally.
Signed-off-by: Jason Lowe-Power <jason@lowepower.com>
This function returns the GPUs (for now, possibly other devices in the
future). It needs to be in the abstract board so the GPU-specific cache
hierarchies can be used with non-GPU boards.
Signed-off-by: Jason Lowe-Power <jason@lowepower.com>
This prepends loading the GPU drivers to anything passed in via the
readfile_contents. Note that if the user sets a specific readfile via a
file they will be responsible for loading the driver
Signed-off-by: Jason Lowe-Power <jason@lowepower.com>
This change separates the instantiation of gpu memory from
instantiatiing the gpu cache. Prior to this change, the gpu
cache instantiated the memories for the gpu by receiving number
of channels as a parameter. With this change, the gpu memory
should be constructed outside the gpu, without being added as a
child to any other object, and passed to the constructor of
the gpu.
This change adds a new method to AbstractMemorySystem to allow
getting its objects of the class MemInterface. This is useful
when certain other classes require a list of MemInterface objects
to create physical memory. In addition, ChanneledMemory and
HighBandwidthMemory implement this function.
