This commit enhances the Translation Lookaside Buffer (TLB) implementation
in the ARM Native Code Execution (NCE) component to increase stability,
particularly on Android devices. The changes prioritize robustness and
error recovery over performance optimizations.
Key improvements:
- Replace set-associative TLB with a simpler linear search implementation
- Implement a basic LRU replacement policy for TLB entries
- Add validation checks for memory addresses before TLB insertion
- Ensure proper page alignment for guest and host addresses
- Enhance alignment fault handling with instruction skipping as fallback
- Add comprehensive debug logging for memory access errors
- Improve error recovery in guest memory access scenarios
These changes should significantly reduce crashes during emulation on
Android devices by gracefully handling memory access edge cases that
previously resulted in hard crashes.
Co-Authored-By: Camille LaVey <camillelavey@citron-emu.org>
Signed-off-by: Zephyron <zephyron@citron-emu.org>
Adds a software TLB cache to improve memory access performance in the NCE
(Native Code Execution) system. Key changes include:
- Implement set-associative TLB with 64 sets and 8 ways
- Add TLB lookup before memory access in HandleGuestAccessFault
- Implement LRU replacement policy with access frequency consideration
- Add thread context caching to reduce overhead
- Add proper synchronization with mutex locks
- Add helper functions for TLB management (lookup, insert, invalidate)
This change should improve performance by reducing redundant memory
translations and providing faster access to frequently used pages.
Update fmt library to version 11.0.2 and make necessary adjustments:
- Replace fmt/format.h includes with fmt/ranges.h
- Add const qualifiers to formatter::format functions
- Update CMake to require fmt version 11
Additional dependency updates:
- Update SDL2 bundled version from 2.28.2 to 2.32.0
- Update catch2 to version 3.8.0
- Update vcpkg baseline to c82f74667287d3dc386bce81e44964370c91a289
Two main changes in this commit:
1. Replace NVIDIA-specific GL_NV_gpu_shader5 extension with the more widely
supported GL_EXT_shader_explicit_arithmetic_types_float16 in the scaleforce
shader. This improves compatibility across different GPU vendors.
2. Refactor ARM32 exception handling:
- Restructure exception cases for better readability
- Update exception handling to match current Dynarmic API
- Fix indentation in switch statement
- Remove AccessViolation case as it's no longer supported in current API
These changes improve shader compatibility and align the exception handling
with the current Dynarmic implementation.
- Update vcpkg baseline to a42af01b72c28a8e1d7b48107b33e4f286a55ef6
- Add SPIRV-Tools and SPIRV-Headers as submodules
- Update Vulkan-related submodules to latest stable versions
- Improve memory access error handling:
- Add specific handling for null pointer accesses in ARM32 emulation
- Return 0 for null pointer reads instead of undefined behavior
- Silently ignore writes to null pointers
- Add more detailed error messages distinguishing between null pointer
access and other unmapped memory errors
- Treat addresses below 0x1000 as potential null pointer accesses
These changes should provide more graceful handling of null pointer
accesses and improve stability when running games that attempt invalid
memory operations.
- In commit b3facaa6bb, the copyright header was
updated to include "Citron Homebrew Project" across multiple files, regardless
of whether any contributions were made.
- This commit removes the incorrect attribution and reverts the copyright header
to its previous state.
- Copyright attribution should only be added when meaningful contributions have
been made to the file.
- This commit ensures proper compliance with copyright standards and maintains
correct attribution to the respective contributors.
- Special thanks to Tachi for pointing out the need for these corrections and
ensuring that proper attribution practices are followed.
Adds logic to track and skip duplicate consecutive addresses when logging
backtraces. This improves log readability by removing redundant entries
that point to the same address in sequence.
There are still some other issues not addressed here, but it's a start.
Workarounds for false-positive reports:
- `RasterizerAccelerated`: Put a gigantic array behind a `unique_ptr`,
because UBSan has a [hardcoded limit](https://stackoverflow.com/questions/64531383/c-runtime-error-using-fsanitize-undefined-object-has-a-possibly-invalid-vp)
of how big it thinks objects can be, specifically when dealing with
offset-to-top values used with multiple inheritance. Hopefully this
doesn't have a performance impact.
- `QueryCacheBase::QueryCacheBase`: Avoid an operation that UBSan thinks
is UB even though it at least arguably isn't. See the link in the
comment for more information.
Fixes for correct reports:
- `PageTable`, `Memory`: Use `uintptr_t` values instead of pointers to
avoid UB from pointer overflow (when pointer arithmetic wraps around
the address space).
- `KScheduler::Reload`: `thread->GetOwnerProcess()` can be `nullptr`;
avoid calling methods on it in this case. (The existing code returns
a garbage reference to a field, which is then passed into
`LoadWatchpointArray`, and apparently it's never used, so it's
harmless in practice but still triggers UBSan.)
- `KAutoObject::Close`: This function calls `this->Destroy()`, which
overwrites the beginning of the object with junk (specifically a free
list pointer). Then it calls `this->UnregisterWithKernel()`. UBSan
complains about a type mismatch because the vtable has been
overwritten, and I believe this is indeed UB. `UnregisterWithKernel`
also loads `m_kernel` from the 'freed' object, which seems to be
technically safe (the overwriting doesn't extend as far as that
field), but seems dubious. Switch to a `static` method and load
`m_kernel` in advance.
