Viewport tags never expire; a freed resource can be passed to the RR runtime

Open
#74 0 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
4/5
Estimated time
1-2 days
Newbie friendliness
35/100
Issue type
Bug
Clarity
Mostly clear
Activity status
Active
Tech stack
cpp

Research direction

Start with store_tags in clean/rr_f/proxy/src/proxy_core.cpp (around lines 548-594), then lookup (around 811) and the evaluate path where describe_resource is called (around 1133). Also check free_resources (around 935) and the matching RrProbe code in probe.cpp. Done means the stage 1 erase rules are in place and a test in test_proxy_core using the fake runtime shows the stale hit-distance pointer never reaches describe_resource or evaluate.

Written by the indexing model from the issue text.

Description

area: ray-reconstruction bug

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

What happens

RrProxy::store_tags (clean/rr_f/proxy/src/proxy_core.cpp:548-594) copies t.resource->native into the stored tag and t.lifetime into s.lifetime. It takes no reference. Only a lifetime-0 tag of a type the runtime reads gets a runtime-owned copy (proxy_core.cpp:574-591).

Entries in viewport_tags_ are removed only by replacement or an explicit null tag (proxy_core.cpp:554-560), by shutdown (proxy_core.cpp:374-375) and by the destructor (proxy_core.cpp:195). lookup falls back to viewport_tags_ with no lifetime or frame check (proxy_core.cpp:811-812). The stored lifetime is never consulted for expiry. new_frame_token only records the index (proxy_core.cpp:708). free_resources destroys the instance and the viewport only (proxy_core.cpp:935).

In the evaluate routing, optional inputs (specular motion, hit distance, disocclusion, guides, layer, opacity, alpha) are used purely because lookup succeeds. The stale pointer is then passed to describe_resource (proxy_core.cpp:1133) and on to the runtime evaluate.

The probe's own RrProbe has the same structure (probe.cpp:497-534).

When it can happen

Needs a game that tags an optional input in some frames and then stops tagging it without tagging null, and frees or recycles the resource. Or a game that frees the RR viewport and reuses its id with a different input set. A game that tags its full set every frame (the common case) or untags with null is not affected.

Plausible triggers: toggling an in-game option that removes an optional guide (depth of field, transparency layer, alpha), or a resolution change that reallocates targets while only the required set is tagged again.

Result: describe_resource or evaluate on a released ID3D12Resource (a crash), or reading an unrelated reallocated resource (garbage in the RR output). Medium severity: a crash is possible, but it needs an uncommon tagging pattern. Related: #63 (RR heaps released at destroy).

To do

Stage 1, now:

  • In free_resources, erase (and release_tag) every viewport_tags_ entry whose key's first element is the freed viewport id.
  • After an RR evaluate, successful or not, erase the viewport_tags_ entries with lifetime == WG_LIFE_UNTIL_EVALUATE for that viewport.

Stage 2, later:

  • Optionally stamp tags with a frame generation and expire lifetime-1 tags at the next frame token. Do this only with evidence from probe logs of real games that they re-tag every frame.

Cautions: a game that tags persistent targets once and relies on the original interface keeping them would lose inputs after the lifetime-1 change. That breaks the contract but may be tolerated by the original. Keep a transient copy alive until the GPU has finished; that is already the runtime's job via release_tag.

Test

CPU-only, in test_proxy_core with the fake runtime.

  • Tag hit-distance (lifetime 2) plus the required set, evaluate. Next frame tag only the required set, evaluate. Assert the fake runtime's describe_resource and evaluate never saw the hit-distance pointer the second time.
  • Free the viewport, allocate the same id again. Assert no old tag reaches the runtime.
Dominant language
C++
Stars
12
Forks
2
Avg merge
3h 10m
Merged PRs (30d)
171

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from phantoslayer21/wronggpu

All issues in phantoslayer21/wronggpu

Similar issues

More C++ issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.