RR silently not used when the game interface is already loaded (delay-load imports)

未关闭
#77 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
4/5
预计耗时
3-5 天
新手友好度
42/100
Issue 类型
缺陷
描述清晰度
基本清楚
活跃度
活跃
技术栈
cpp

调研方向

Start in clean/rr_f/proxy/src/hook_dll.cpp at the delay-import lookup inside if (!g_interposer) (around lines 1050-1065), then the module loop and if (delay) patch block (around lines 1105-1120). Read the test header clean/rr_f/proxy/tests/test_hook_dll.cpp to see the WG_TEST_DELAYLOAD setup. Done means the RR delay slots point into the plugin when the interface is already loaded, the host route is requested when nothing was patched, and the test passes.

由索引模型根据 Issue 内容生成。

描述

area: ray-reconstruction bug needs-hardware-test

This was found by reading the code. It has not been reproduced.

What happens

In do_hook (clean/rr_f/proxy/src/hook_dll.cpp), the delay-import descriptor is looked up only inside if (!g_interposer) (hook_dll.cpp:1050-1053). If GetModuleHandleW already finds the interface module, delay stays null. The log then says "already loaded via import table" (hook_dll.cpp:1065), even when the exe imports it by delay-load or another module loaded it first.

The module loop patches ordinary import tables only (hook_dll.cpp:1105-1115). The only delay-IAT patch is under if (delay) (hook_dll.cpp:1116-1120). g_wait_for_host is set only when the interface is absent and not linked (hook_dll.cpp:1057), so the request for the host's route (hook_dll.cpp:1346) is not made either.

Result: the exe's delay-load slots for the interface are never patched, no host route is requested, and the game's RR calls go to the original interface. RR is silently not used. The log is misleading. There is no crash.

When it can happen

A game that delay-imports the interface, where the delay load has already fired before InitializeASI (the host loads the plugin after the game has started), or where another module (host, overlay) loaded the interface first. This timing is ordinary, not exotic. The test header (clean/rr_f/proxy/tests/test_hook_dll.cpp) covers only the "not loaded yet" branch and says the host loads the plugin after the interface is already loaded for static importers. What is lost: the RR feature on affected delay-load games, with no visible error. Medium severity.

To do

  • Always compute delay = find_delay_import(GetModuleHandleW(nullptr), kWgInterposerFile). If present, call patch_module_delay_imports whether or not the module was already loaded.
  • Keep "we loaded it ourselves" as a separate bool for the adopt_unseen_init decision (hook_dll.cpp:1088-1095). If the interface was already loaded, init may have run, so adopt.
  • If the total patched count is 0, set g_wait_for_host so the host route is still requested.
  • Fix the log line so it does not claim an import-table load when it was not.
  • Caution: patching an already-resolved delay slot is fine, the same as an IAT slot. Do not leave the adopt decision keyed on delay, or an already-initialised interface would be treated as fresh.

Test

CPU-only. In test_hook_dll, built with WG_TEST_DELAYLOAD: call one non-RR interface export (forcing the delay load), or LoadLibrary the interface before loading the plugin. Assert the RR delay slots now point into the plugin and the RR path reaches the fake runtime. A run in a real delay-load game would confirm it (see #36 for testing in a game).

主要语言
C++
星标
12
派生
2
平均合并
3 小时 10 分钟
30 天内合并 PR
171

环境准备

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

phantoslayer21/wronggpu 的其他 Issue

查看 phantoslayer21/wronggpu 的全部 Issue

相似的 Issue

更多 C++ Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。