Jmq/rwlock device state#80
Merged
Merged
Conversation
Continues removing static muts. DEVICE_STATE was a `*mut DeviceState` guarded by a separate marker `RwLock<()>` (DEVICE_STATE_LOCK), which was used inconsistently and often skipped entirely. This collapses the two into a single `RwLock<DeviceStatePtr>`, where DeviceStatePtr is a Send/Sync wrapper around the raw pointer, so every read/write goes through the same lock. API changes in the device_state crate: - `dev_state()` and `dev_state_d3d11_nolock()` are gone. Both used to hand out a `&'static mut`, which can't be reconciled with a guard lifetime, so every call site was migrated. - `dev_state_write()` / `dev_state_read()` return `Option<(guard, &/&mut DeviceState)>`. - `dev_state_d3d11_write()` and `dev_state_d3d11_read()` keep their shape, but `_read` now returns `&` instead of `&mut` (the previous `_read` was unsound). - All accessors fail safely on lock poisoning by logging and returning None rather than panicking, so a poison can't take the game down. Re-entrancy: `std::sync::RwLock` is not reentrant, so several call sites had to be reworked to drop the guard before invoking real D3D fns (which can re-enter our hooks on the same thread). Notably: - D3D9 `hook_present`, `hook_release`, `hook_set_texture`, `hook_set_stream_source`, `hook_reset`, `hook_draw_indexed_primitive`, `hook_create_device`, `hook_CreateTexture`, `hook_UpdateTexture`, and `create_d3d9` now extract the real fn pointer under a brief read guard, drop the guard, then invoke. The release path also threads `ref_count` updates through scoped write guards. - `HookDirect3D11Context` and friends now derive Copy so `get_hook_context` returns a copy and callers don't need to hold the dev_state lock to invoke the real fn pointers. - DX11 `hook_draw_indexed` was restructured so dev_state guards are acquired only briefly: the geom check / vb_state snapshot under a read guard, mod work outside any guard, and metrics under a final write guard. `ensure_vb_checksum_dx11`, `hook_snapshot::take`, and `with_dev_ptr` are no longer called with the lock held. - `init_device_state_once` now mutates the pointer through the write guard rather than via static mut. Tests serialize on a separate `TEST_DEV_STATE_LOCK` because the new init path takes the dev_state write lock internally.
JMQ Note: This merge includes the device state RWLock. Previously it was just a static mut and there was the potential for undefined behavior since multiple parts of the code were potentially creating &mut's to it at the same time. Inside an RWLock, this can't happen. Performance testing was not completely definitive but I couldn't pinpoint any real slowdown from this. For future work, possibly consider parking lot RwLock as an optimization - but I did try it briefly and did not see any real speedup. This required a conflict resolution between the refill code which claude describes hereafter: The conflict was between this branch's RwLock-based dev_state access pattern and master's new "needs refill for layout" mod reload logic (#77). Resolution preserves the RwLock pattern (lock acquired/dropped in narrow scopes) while integrating the refill logic: we precompute the current input layout's semantic mask via a brief read lock before calling check_and_render_mod, then pass scalars into the closure so it no longer needs the dev_state lock. needs_refill_for_layout was refactored to accept an Option<SemanticMask> instead of taking a reference to the layouts hashmap, so callers can look up the mask once under their read guard rather than holding the lock across the closure. The post_mod_check NotRenderedButLoadRequested arm now also calls mod_load::reset_for_reload from master, while keeping the rwlock pattern of reading current_input_layout via a brief dev_state guard before taking the LOADED_MODS lock.
jmquigs
force-pushed
the
jmq/rwlock-device-state
branch
from
ae90dc2 to
3a7b1b5
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.