PR 0010 PGRAPH reject oversized inline packets atomically

PR #10 — PGRAPH: reject oversized inline packets atomically

PR #10

Status

• GitHub state: Merged
• Disposition at review: Merged
• Current head at review: 659ad30be0488e85bea0df9514a5e52a31fcf290 (feature/eng-2026-433-s-pr09-pfifo-bulk-uc037deed)
• Reviewed head: 659ad30be0488e85bea0df9514a5e52a31fcf290
• Last reviewed: 2026-09-14 UTC
• Base: main at db0905eabd5a1fd0ee1a8e8eb81585db104e256e
• Subsystem: NV2A / PGRAPH

Review freshness: this page records the final/known PR head at review time. If the PR is reopened or rewritten, re-review before relying on implementation details.

At a glance

• What it changes: PGRAPH: reject oversized inline packets atomically
• Why it was attempted: Observed problem: Oversized ARRAY ELEMENT16 , ARRAY ELEMENT32 , and INLINE ARRAY packets could reach assertion-backed expansion or mutation paths before the complete packet capacity was known. Profiling evidence: The earlier valid-packet work showed that bulk inline handling can reduce repeated PFIFO dispatch/bookkeeping. The current-main reconstruction therefore had two requirements: preserve that accepted-packet path and restore whole-packet preflight b…
• Current/final result: Merged: atomic packet-boundary tests, scalar/bulk equivalence, validation, and accepted-packet performance gates passed.
• Lesson: Validate the complete inline packet capacity before expanding pending draw work or mutating an inline destination; oversized packets must be rejected atomically.

Code impact

Implementation file	Δ	Why it matters
hw/xbox/nv2a/pgraph/inline-elements.h	+27/-6	PGRAPH inline-packet capacity and atomic-mutation contract.
hw/xbox/nv2a/pgraph/pgraph.c	+50/-0	Shared PGRAPH guest-state/method handling and renderer boundary.
tests/xbox/pgraph-inline-elements/pgraph-inline-elements-test.c	+25/-0	Focused/regression validation for the implementation contract.

Flow

flowchart LR
    subgraph Before
        A[Guest inline packet] --> B[Begin expansion/mutation] --> C[Discover capacity problem] --> D[Partial state possible]
    end
    subgraph PR_10[PR #10]
        E[Guest inline packet] --> F[PR #10 capacity preflight] --> G[Atomic accept or reject] --> H[Consistent PGRAPH state]
    end

Loading

What changed

Restore pending-draw-aware packet planning. - Preflight complete inline packets before handler dispatch/mutation. - Account for ELEMENT16's two values per word. - Consume malformed dispatch units through a controlled guest-error path. - Preserve existing valid bulk/scalar loops and add focused boundary/equivalence coverage. Main code path: hw/xbox/nv2a/pgraph/pgraph.c , inline-elements.h , focused PFIFO tests Commits: merged candidate 659ad30be0488e85bea0df9514a5e52a31fcf290

Result

Merged: atomic packet-boundary tests, scalar/bulk equivalence, validation, and accepted-packet performance gates passed.

Lesson learned

Validate the complete inline packet capacity before expanding pending draw work or mutating an inline destination; oversized packets must be rejected atomically.

Avoid repeating

Do not begin ARRAY_ELEMENT/INLINE_ARRAY mutation and then discover the packet is oversized; partial mutation leaves PGRAPH state inconsistent.

Docs / evidence

Pointers only; the detailed evidence remains in its original location.

• tests/xbox/pgraph-inline-elements/
• https://github.com/Mainkill1/xemu-perf-tests/pull/16
• https://github.com/Mainkill1/xemu-perf-tests/pull/18
• https://github.com/Mainkill1/xemu-perf-tests/pull/17
• https://github.com/Mainkill1/xemu/blob/13cb6d7384b26490cdec2ccf08fc1d607efc426a/evidence/eng-2026-433-pr10-pfifo-hardening/REPORT.md

Related architecture

• NV2A-Graphics-Architecture
• NV2A Graphics Architecture